xref: /freebsd/sys/netinet6/mld6.c (revision 037479ff5ee18977b1c48e1e59770aad2f200a5a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
66  */
67 
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
70 
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
84 #include <sys/ktr.h>
85 
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90 
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
100 
101 #include <security/mac/mac_framework.h>
102 
103 #ifndef KTR_MLD
104 #define KTR_MLD KTR_INET6
105 #endif
106 
107 static struct mld_ifsoftc *
108 		mli_alloc_locked(struct ifnet *);
109 static void	mli_delete_locked(const struct ifnet *);
110 static void	mld_dispatch_packet(struct mbuf *);
111 static void	mld_dispatch_queue(struct mbufq *, int);
112 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
113 static void	mld_fasttimo_vnet(void);
114 static int	mld_handle_state_change(struct in6_multi *,
115 		    struct mld_ifsoftc *);
116 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
117 		    const int);
118 #ifdef KTR
119 static char *	mld_rec_type_to_str(const int);
120 #endif
121 static void	mld_set_version(struct mld_ifsoftc *, const int);
122 static void	mld_slowtimo_vnet(void);
123 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
126 		    /*const*/ struct mld_hdr *);
127 static void	mld_v1_process_group_timer(struct in6_multi_head *,
128 		    struct in6_multi *);
129 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
130 static int	mld_v1_transmit_report(struct in6_multi *, const int);
131 static void	mld_v1_update_group(struct in6_multi *, const int);
132 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
133 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
134 static struct mbuf *
135 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
136 static int	mld_v2_enqueue_filter_change(struct mbufq *,
137 		    struct in6_multi *);
138 static int	mld_v2_enqueue_group_record(struct mbufq *,
139 		    struct in6_multi *, const int, const int, const int,
140 		    const int);
141 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
142 		    struct mbuf *, const int, const int);
143 static int	mld_v2_merge_state_changes(struct in6_multi *,
144 		    struct mbufq *);
145 static void	mld_v2_process_group_timers(struct in6_multi_head *,
146 		    struct mbufq *, struct mbufq *,
147 		    struct in6_multi *, const int);
148 static int	mld_v2_process_group_query(struct in6_multi *,
149 		    struct mld_ifsoftc *mli, int, struct mbuf *, const int);
150 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
151 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
152 
153 /*
154  * Normative references: RFC 2710, RFC 3590, RFC 3810.
155  *
156  * Locking:
157  *  * The MLD subsystem lock ends up being system-wide for the moment,
158  *    but could be per-VIMAGE later on.
159  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
160  *    Any may be taken independently; if any are held at the same
161  *    time, the above lock order must be followed.
162  *  * IN6_MULTI_LOCK covers in_multi.
163  *  * MLD_LOCK covers per-link state and any global variables in this file.
164  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
165  *    per-link state iterators.
166  *
167  *  XXX LOR PREVENTION
168  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
169  *  will not accept an ifp; it wants an embedded scope ID, unlike
170  *  ip_output(), which happily takes the ifp given to it. The embedded
171  *  scope ID is only used by MLD to select the outgoing interface.
172  *
173  *  During interface attach and detach, MLD will take MLD_LOCK *after*
174  *  the IF_AFDATA_LOCK.
175  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
176  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
177  *  dispatch could work around this, but we'd rather not do that, as it
178  *  can introduce other races.
179  *
180  *  As such, we exploit the fact that the scope ID is just the interface
181  *  index, and embed it in the IPv6 destination address accordingly.
182  *  This is potentially NOT VALID for MLDv1 reports, as they
183  *  are always sent to the multicast group itself; as MLDv2
184  *  reports are always sent to ff02::16, this is not an issue
185  *  when MLDv2 is in use.
186  *
187  *  This does not however eliminate the LOR when ip6_output() itself
188  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
189  *  trigger a LOR warning in WITNESS when the ifnet is detached.
190  *
191  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
192  *  how it's used across the network stack. Here we're simply exploiting
193  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
194  *
195  * VIMAGE:
196  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
197  *    to a vnet in ifp->if_vnet.
198  */
199 static struct mtx		 mld_mtx;
200 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
201 
202 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
203 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
204 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
205 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
206 
207 /*
208  * VIMAGE-wide globals.
209  */
210 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
211 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
212 VNET_DEFINE_STATIC(int, interface_timers_running6);
213 VNET_DEFINE_STATIC(int, state_change_timers_running6);
214 VNET_DEFINE_STATIC(int, current_state_timers_running6);
215 
216 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
217 #define	V_mli_head			VNET(mli_head)
218 #define	V_interface_timers_running6	VNET(interface_timers_running6)
219 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
220 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
221 
222 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
223 
224 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
225     "IPv6 Multicast Listener Discovery");
226 
227 /*
228  * Virtualized sysctls.
229  */
230 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
231     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
232     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
233     "Rate limit for MLDv2 Group-and-Source queries in seconds");
234 
235 /*
236  * Non-virtualized sysctls.
237  */
238 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
239     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
240     "Per-interface MLDv2 state");
241 
242 static int	mld_v1enable = 1;
243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
244     &mld_v1enable, 0, "Enable fallback to MLDv1");
245 
246 static int	mld_use_allow = 1;
247 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
248     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
249 
250 /*
251  * Packed Router Alert option structure declaration.
252  */
253 struct mld_raopt {
254 	struct ip6_hbh		hbh;
255 	struct ip6_opt		pad;
256 	struct ip6_opt_router	ra;
257 } __packed;
258 
259 /*
260  * Router Alert hop-by-hop option header.
261  */
262 static struct mld_raopt mld_ra = {
263 	.hbh = { 0, 0 },
264 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
265 	.ra = {
266 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
267 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
268 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
269 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
270 	}
271 };
272 static struct ip6_pktopts mld_po;
273 
274 static __inline void
275 mld_save_context(struct mbuf *m, struct ifnet *ifp)
276 {
277 
278 #ifdef VIMAGE
279 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
280 #endif /* VIMAGE */
281 	m->m_pkthdr.flowid = ifp->if_index;
282 }
283 
284 static __inline void
285 mld_scrub_context(struct mbuf *m)
286 {
287 
288 	m->m_pkthdr.PH_loc.ptr = NULL;
289 	m->m_pkthdr.flowid = 0;
290 }
291 
292 /*
293  * Restore context from a queued output chain.
294  * Return saved ifindex.
295  *
296  * VIMAGE: The assertion is there to make sure that we
297  * actually called CURVNET_SET() with what's in the mbuf chain.
298  */
299 static __inline uint32_t
300 mld_restore_context(struct mbuf *m)
301 {
302 
303 #if defined(VIMAGE) && defined(INVARIANTS)
304 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
305 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
306 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
307 #endif
308 	return (m->m_pkthdr.flowid);
309 }
310 
311 /*
312  * Retrieve or set threshold between group-source queries in seconds.
313  *
314  * VIMAGE: Assume curvnet set by caller.
315  * SMPng: NOTE: Serialized by MLD lock.
316  */
317 static int
318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
319 {
320 	int error;
321 	int i;
322 
323 	error = sysctl_wire_old_buffer(req, sizeof(int));
324 	if (error)
325 		return (error);
326 
327 	MLD_LOCK();
328 
329 	i = V_mld_gsrdelay.tv_sec;
330 
331 	error = sysctl_handle_int(oidp, &i, 0, req);
332 	if (error || !req->newptr)
333 		goto out_locked;
334 
335 	if (i < -1 || i >= 60) {
336 		error = EINVAL;
337 		goto out_locked;
338 	}
339 
340 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
341 	     V_mld_gsrdelay.tv_sec, i);
342 	V_mld_gsrdelay.tv_sec = i;
343 
344 out_locked:
345 	MLD_UNLOCK();
346 	return (error);
347 }
348 
349 /*
350  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
351  * For use by ifmcstat(8).
352  *
353  * SMPng: NOTE: Does an unlocked ifindex space read.
354  * VIMAGE: Assume curvnet set by caller. The node handler itself
355  * is not directly virtualized.
356  */
357 static int
358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
359 {
360 	int			*name;
361 	int			 error;
362 	u_int			 namelen;
363 	struct ifnet		*ifp;
364 	struct mld_ifsoftc	*mli;
365 
366 	name = (int *)arg1;
367 	namelen = arg2;
368 
369 	if (req->newptr != NULL)
370 		return (EPERM);
371 
372 	if (namelen != 1)
373 		return (EINVAL);
374 
375 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
376 	if (error)
377 		return (error);
378 
379 	IN6_MULTI_LOCK();
380 	IN6_MULTI_LIST_LOCK();
381 	MLD_LOCK();
382 
383 	if (name[0] <= 0 || name[0] > V_if_index) {
384 		error = ENOENT;
385 		goto out_locked;
386 	}
387 
388 	error = ENOENT;
389 
390 	ifp = ifnet_byindex(name[0]);
391 	if (ifp == NULL)
392 		goto out_locked;
393 
394 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
395 		if (ifp == mli->mli_ifp) {
396 			struct mld_ifinfo info;
397 
398 			info.mli_version = mli->mli_version;
399 			info.mli_v1_timer = mli->mli_v1_timer;
400 			info.mli_v2_timer = mli->mli_v2_timer;
401 			info.mli_flags = mli->mli_flags;
402 			info.mli_rv = mli->mli_rv;
403 			info.mli_qi = mli->mli_qi;
404 			info.mli_qri = mli->mli_qri;
405 			info.mli_uri = mli->mli_uri;
406 			error = SYSCTL_OUT(req, &info, sizeof(info));
407 			break;
408 		}
409 	}
410 
411 out_locked:
412 	MLD_UNLOCK();
413 	IN6_MULTI_LIST_UNLOCK();
414 	IN6_MULTI_UNLOCK();
415 	return (error);
416 }
417 
418 /*
419  * Dispatch an entire queue of pending packet chains.
420  * VIMAGE: Assumes the vnet pointer has been set.
421  */
422 static void
423 mld_dispatch_queue(struct mbufq *mq, int limit)
424 {
425 	struct mbuf *m;
426 
427 	while ((m = mbufq_dequeue(mq)) != NULL) {
428 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
429 		mld_dispatch_packet(m);
430 		if (--limit == 0)
431 			break;
432 	}
433 }
434 
435 /*
436  * Filter outgoing MLD report state by group.
437  *
438  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
439  * and node-local addresses. However, kernel and socket consumers
440  * always embed the KAME scope ID in the address provided, so strip it
441  * when performing comparison.
442  * Note: This is not the same as the *multicast* scope.
443  *
444  * Return zero if the given group is one for which MLD reports
445  * should be suppressed, or non-zero if reports should be issued.
446  */
447 static __inline int
448 mld_is_addr_reported(const struct in6_addr *addr)
449 {
450 
451 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
452 
453 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
454 		return (0);
455 
456 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
457 		struct in6_addr tmp = *addr;
458 		in6_clearscope(&tmp);
459 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
460 			return (0);
461 	}
462 
463 	return (1);
464 }
465 
466 /*
467  * Attach MLD when PF_INET6 is attached to an interface.
468  *
469  * SMPng: Normally called with IF_AFDATA_LOCK held.
470  */
471 struct mld_ifsoftc *
472 mld_domifattach(struct ifnet *ifp)
473 {
474 	struct mld_ifsoftc *mli;
475 
476 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
477 	    __func__, ifp, if_name(ifp));
478 
479 	MLD_LOCK();
480 
481 	mli = mli_alloc_locked(ifp);
482 	if (!(ifp->if_flags & IFF_MULTICAST))
483 		mli->mli_flags |= MLIF_SILENT;
484 	if (mld_use_allow)
485 		mli->mli_flags |= MLIF_USEALLOW;
486 
487 	MLD_UNLOCK();
488 
489 	return (mli);
490 }
491 
492 /*
493  * VIMAGE: assume curvnet set by caller.
494  */
495 static struct mld_ifsoftc *
496 mli_alloc_locked(/*const*/ struct ifnet *ifp)
497 {
498 	struct mld_ifsoftc *mli;
499 
500 	MLD_LOCK_ASSERT();
501 
502 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
503 	if (mli == NULL)
504 		goto out;
505 
506 	mli->mli_ifp = ifp;
507 	mli->mli_version = MLD_VERSION_2;
508 	mli->mli_flags = 0;
509 	mli->mli_rv = MLD_RV_INIT;
510 	mli->mli_qi = MLD_QI_INIT;
511 	mli->mli_qri = MLD_QRI_INIT;
512 	mli->mli_uri = MLD_URI_INIT;
513 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
514 
515 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
516 
517 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
518 	     ifp, if_name(ifp));
519 
520 out:
521 	return (mli);
522 }
523 
524 /*
525  * Hook for ifdetach.
526  *
527  * NOTE: Some finalization tasks need to run before the protocol domain
528  * is detached, but also before the link layer does its cleanup.
529  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
530  *
531  * SMPng: Caller must hold IN6_MULTI_LOCK().
532  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
533  * XXX This routine is also bitten by unlocked ifma_protospec access.
534  */
535 void
536 mld_ifdetach(struct ifnet *ifp)
537 {
538 	struct mld_ifsoftc	*mli;
539 	struct ifmultiaddr	*ifma, *next;
540 	struct in6_multi	*inm;
541 	struct in6_multi_head inmh;
542 
543 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
544 	    if_name(ifp));
545 
546 	SLIST_INIT(&inmh);
547 	IN6_MULTI_LIST_LOCK_ASSERT();
548 	MLD_LOCK();
549 
550 	mli = MLD_IFINFO(ifp);
551 	if (mli->mli_version == MLD_VERSION_2) {
552 		IF_ADDR_WLOCK(ifp);
553 	restart:
554 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
555 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
556 			    ifma->ifma_protospec == NULL)
557 				continue;
558 			inm = (struct in6_multi *)ifma->ifma_protospec;
559 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
560 				in6m_disconnect(inm);
561 				in6m_rele_locked(&inmh, inm);
562 				ifma->ifma_protospec = NULL;
563 			}
564 			in6m_clear_recorded(inm);
565 			if (__predict_false(ifma6_restart)) {
566 				ifma6_restart = false;
567 				goto restart;
568 			}
569 		}
570 		IF_ADDR_WUNLOCK(ifp);
571 	}
572 
573 	MLD_UNLOCK();
574 	in6m_release_list_deferred(&inmh);
575 }
576 
577 /*
578  * Hook for domifdetach.
579  * Runs after link-layer cleanup; free MLD state.
580  *
581  * SMPng: Normally called with IF_AFDATA_LOCK held.
582  */
583 void
584 mld_domifdetach(struct ifnet *ifp)
585 {
586 
587 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
588 	    __func__, ifp, if_name(ifp));
589 
590 	MLD_LOCK();
591 	mli_delete_locked(ifp);
592 	MLD_UNLOCK();
593 }
594 
595 static void
596 mli_delete_locked(const struct ifnet *ifp)
597 {
598 	struct mld_ifsoftc *mli, *tmli;
599 
600 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
601 	    __func__, ifp, if_name(ifp));
602 
603 	MLD_LOCK_ASSERT();
604 
605 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
606 		if (mli->mli_ifp == ifp) {
607 			/*
608 			 * Free deferred General Query responses.
609 			 */
610 			mbufq_drain(&mli->mli_gq);
611 
612 			LIST_REMOVE(mli, mli_link);
613 
614 			free(mli, M_MLD);
615 			return;
616 		}
617 	}
618 }
619 
620 /*
621  * Process a received MLDv1 general or address-specific query.
622  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
623  *
624  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
625  * mld_addr. This is OK as we own the mbuf chain.
626  */
627 static int
628 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
629     /*const*/ struct mld_hdr *mld)
630 {
631 	struct ifmultiaddr	*ifma;
632 	struct mld_ifsoftc	*mli;
633 	struct in6_multi	*inm;
634 	int			 is_general_query;
635 	uint16_t		 timer;
636 #ifdef KTR
637 	char			 ip6tbuf[INET6_ADDRSTRLEN];
638 #endif
639 
640 	is_general_query = 0;
641 
642 	if (!mld_v1enable) {
643 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
644 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
645 		    ifp, if_name(ifp));
646 		return (0);
647 	}
648 
649 	/*
650 	 * RFC3810 Section 6.2: MLD queries must originate from
651 	 * a router's link-local address.
652 	 */
653 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
654 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
655 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
656 		    ifp, if_name(ifp));
657 		return (0);
658 	}
659 
660 	/*
661 	 * Do address field validation upfront before we accept
662 	 * the query.
663 	 */
664 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
665 		/*
666 		 * MLDv1 General Query.
667 		 * If this was not sent to the all-nodes group, ignore it.
668 		 */
669 		struct in6_addr		 dst;
670 
671 		dst = ip6->ip6_dst;
672 		in6_clearscope(&dst);
673 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
674 			return (EINVAL);
675 		is_general_query = 1;
676 	} else {
677 		/*
678 		 * Embed scope ID of receiving interface in MLD query for
679 		 * lookup whilst we don't hold other locks.
680 		 */
681 		in6_setscope(&mld->mld_addr, ifp, NULL);
682 	}
683 
684 	IN6_MULTI_LIST_LOCK();
685 	MLD_LOCK();
686 
687 	/*
688 	 * Switch to MLDv1 host compatibility mode.
689 	 */
690 	mli = MLD_IFINFO(ifp);
691 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
692 	mld_set_version(mli, MLD_VERSION_1);
693 
694 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
695 	if (timer == 0)
696 		timer = 1;
697 
698 	IF_ADDR_RLOCK(ifp);
699 	if (is_general_query) {
700 		/*
701 		 * For each reporting group joined on this
702 		 * interface, kick the report timer.
703 		 */
704 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
705 			 ifp, if_name(ifp));
706 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
707 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
708 			    ifma->ifma_protospec == NULL)
709 				continue;
710 			inm = (struct in6_multi *)ifma->ifma_protospec;
711 			mld_v1_update_group(inm, timer);
712 		}
713 	} else {
714 		/*
715 		 * MLDv1 Group-Specific Query.
716 		 * If this is a group-specific MLDv1 query, we need only
717 		 * look up the single group to process it.
718 		 */
719 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
720 		if (inm != NULL) {
721 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
722 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
723 			    ifp, if_name(ifp));
724 			mld_v1_update_group(inm, timer);
725 		}
726 		/* XXX Clear embedded scope ID as userland won't expect it. */
727 		in6_clearscope(&mld->mld_addr);
728 	}
729 
730 	IF_ADDR_RUNLOCK(ifp);
731 	MLD_UNLOCK();
732 	IN6_MULTI_LIST_UNLOCK();
733 
734 	return (0);
735 }
736 
737 /*
738  * Update the report timer on a group in response to an MLDv1 query.
739  *
740  * If we are becoming the reporting member for this group, start the timer.
741  * If we already are the reporting member for this group, and timer is
742  * below the threshold, reset it.
743  *
744  * We may be updating the group for the first time since we switched
745  * to MLDv2. If we are, then we must clear any recorded source lists,
746  * and transition to REPORTING state; the group timer is overloaded
747  * for group and group-source query responses.
748  *
749  * Unlike MLDv2, the delay per group should be jittered
750  * to avoid bursts of MLDv1 reports.
751  */
752 static void
753 mld_v1_update_group(struct in6_multi *inm, const int timer)
754 {
755 #ifdef KTR
756 	char			 ip6tbuf[INET6_ADDRSTRLEN];
757 #endif
758 
759 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
760 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
761 	    if_name(inm->in6m_ifp), timer);
762 
763 	IN6_MULTI_LIST_LOCK_ASSERT();
764 
765 	switch (inm->in6m_state) {
766 	case MLD_NOT_MEMBER:
767 	case MLD_SILENT_MEMBER:
768 		break;
769 	case MLD_REPORTING_MEMBER:
770 		if (inm->in6m_timer != 0 &&
771 		    inm->in6m_timer <= timer) {
772 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
773 			    "skipping.", __func__);
774 			break;
775 		}
776 		/* FALLTHROUGH */
777 	case MLD_SG_QUERY_PENDING_MEMBER:
778 	case MLD_G_QUERY_PENDING_MEMBER:
779 	case MLD_IDLE_MEMBER:
780 	case MLD_LAZY_MEMBER:
781 	case MLD_AWAKENING_MEMBER:
782 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
783 		inm->in6m_state = MLD_REPORTING_MEMBER;
784 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
785 		V_current_state_timers_running6 = 1;
786 		break;
787 	case MLD_SLEEPING_MEMBER:
788 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
789 		inm->in6m_state = MLD_AWAKENING_MEMBER;
790 		break;
791 	case MLD_LEAVING_MEMBER:
792 		break;
793 	}
794 }
795 
796 /*
797  * Process a received MLDv2 general, group-specific or
798  * group-and-source-specific query.
799  *
800  * Assumes that the query header has been pulled up to sizeof(mldv2_query).
801  *
802  * Return 0 if successful, otherwise an appropriate error code is returned.
803  */
804 static int
805 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
806     struct mbuf *m, const int off, const int icmp6len)
807 {
808 	struct mld_ifsoftc	*mli;
809 	struct mldv2_query	*mld;
810 	struct in6_multi	*inm;
811 	uint32_t		 maxdelay, nsrc, qqi;
812 	int			 is_general_query;
813 	uint16_t		 timer;
814 	uint8_t			 qrv;
815 #ifdef KTR
816 	char			 ip6tbuf[INET6_ADDRSTRLEN];
817 #endif
818 
819 	is_general_query = 0;
820 
821 	/*
822 	 * RFC3810 Section 6.2: MLD queries must originate from
823 	 * a router's link-local address.
824 	 */
825 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
826 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
827 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
828 		    ifp, if_name(ifp));
829 		return (0);
830 	}
831 
832 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
833 
834 	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
835 
836 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
837 	if (maxdelay >= 32768) {
838 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
839 			   (MLD_MRC_EXP(maxdelay) + 3);
840 	}
841 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
842 	if (timer == 0)
843 		timer = 1;
844 
845 	qrv = MLD_QRV(mld->mld_misc);
846 	if (qrv < 2) {
847 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
848 		    qrv, MLD_RV_INIT);
849 		qrv = MLD_RV_INIT;
850 	}
851 
852 	qqi = mld->mld_qqi;
853 	if (qqi >= 128) {
854 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
855 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
856 	}
857 
858 	nsrc = ntohs(mld->mld_numsrc);
859 	if (nsrc > MLD_MAX_GS_SOURCES)
860 		return (EMSGSIZE);
861 	if (icmp6len < sizeof(struct mldv2_query) +
862 	    (nsrc * sizeof(struct in6_addr)))
863 		return (EMSGSIZE);
864 
865 	/*
866 	 * Do further input validation upfront to avoid resetting timers
867 	 * should we need to discard this query.
868 	 */
869 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
870 		/*
871 		 * A general query with a source list has undefined
872 		 * behaviour; discard it.
873 		 */
874 		if (nsrc > 0)
875 			return (EINVAL);
876 		is_general_query = 1;
877 	} else {
878 		/*
879 		 * Embed scope ID of receiving interface in MLD query for
880 		 * lookup whilst we don't hold other locks (due to KAME
881 		 * locking lameness). We own this mbuf chain just now.
882 		 */
883 		in6_setscope(&mld->mld_addr, ifp, NULL);
884 	}
885 
886 	IN6_MULTI_LIST_LOCK();
887 	MLD_LOCK();
888 
889 	mli = MLD_IFINFO(ifp);
890 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
891 
892 	/*
893 	 * Discard the v2 query if we're in Compatibility Mode.
894 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
895 	 * until the Old Version Querier Present timer expires.
896 	 */
897 	if (mli->mli_version != MLD_VERSION_2)
898 		goto out_locked;
899 
900 	mld_set_version(mli, MLD_VERSION_2);
901 	mli->mli_rv = qrv;
902 	mli->mli_qi = qqi;
903 	mli->mli_qri = maxdelay;
904 
905 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
906 	    maxdelay);
907 
908 	if (is_general_query) {
909 		/*
910 		 * MLDv2 General Query.
911 		 *
912 		 * Schedule a current-state report on this ifp for
913 		 * all groups, possibly containing source lists.
914 		 *
915 		 * If there is a pending General Query response
916 		 * scheduled earlier than the selected delay, do
917 		 * not schedule any other reports.
918 		 * Otherwise, reset the interface timer.
919 		 */
920 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
921 		    ifp, if_name(ifp));
922 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
923 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
924 			V_interface_timers_running6 = 1;
925 		}
926 	} else {
927 		/*
928 		 * MLDv2 Group-specific or Group-and-source-specific Query.
929 		 *
930 		 * Group-source-specific queries are throttled on
931 		 * a per-group basis to defeat denial-of-service attempts.
932 		 * Queries for groups we are not a member of on this
933 		 * link are simply ignored.
934 		 */
935 		IF_ADDR_RLOCK(ifp);
936 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
937 		if (inm == NULL) {
938 			IF_ADDR_RUNLOCK(ifp);
939 			goto out_locked;
940 		}
941 		if (nsrc > 0) {
942 			if (!ratecheck(&inm->in6m_lastgsrtv,
943 			    &V_mld_gsrdelay)) {
944 				CTR1(KTR_MLD, "%s: GS query throttled.",
945 				    __func__);
946 				IF_ADDR_RUNLOCK(ifp);
947 				goto out_locked;
948 			}
949 		}
950 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
951 		     ifp, if_name(ifp));
952 		/*
953 		 * If there is a pending General Query response
954 		 * scheduled sooner than the selected delay, no
955 		 * further report need be scheduled.
956 		 * Otherwise, prepare to respond to the
957 		 * group-specific or group-and-source query.
958 		 */
959 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
960 			mld_v2_process_group_query(inm, mli, timer, m, off);
961 
962 		/* XXX Clear embedded scope ID as userland won't expect it. */
963 		in6_clearscope(&mld->mld_addr);
964 		IF_ADDR_RUNLOCK(ifp);
965 	}
966 
967 out_locked:
968 	MLD_UNLOCK();
969 	IN6_MULTI_LIST_UNLOCK();
970 
971 	return (0);
972 }
973 
974 /*
975  * Process a received MLDv2 group-specific or group-and-source-specific
976  * query.
977  * Return <0 if any error occurred. Currently this is ignored.
978  */
979 static int
980 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
981     int timer, struct mbuf *m0, const int off)
982 {
983 	struct mldv2_query	*mld;
984 	int			 retval;
985 	uint16_t		 nsrc;
986 
987 	IN6_MULTI_LIST_LOCK_ASSERT();
988 	MLD_LOCK_ASSERT();
989 
990 	retval = 0;
991 	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
992 
993 	switch (inm->in6m_state) {
994 	case MLD_NOT_MEMBER:
995 	case MLD_SILENT_MEMBER:
996 	case MLD_SLEEPING_MEMBER:
997 	case MLD_LAZY_MEMBER:
998 	case MLD_AWAKENING_MEMBER:
999 	case MLD_IDLE_MEMBER:
1000 	case MLD_LEAVING_MEMBER:
1001 		return (retval);
1002 		break;
1003 	case MLD_REPORTING_MEMBER:
1004 	case MLD_G_QUERY_PENDING_MEMBER:
1005 	case MLD_SG_QUERY_PENDING_MEMBER:
1006 		break;
1007 	}
1008 
1009 	nsrc = ntohs(mld->mld_numsrc);
1010 
1011 	/*
1012 	 * Deal with group-specific queries upfront.
1013 	 * If any group query is already pending, purge any recorded
1014 	 * source-list state if it exists, and schedule a query response
1015 	 * for this group-specific query.
1016 	 */
1017 	if (nsrc == 0) {
1018 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1019 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1020 			in6m_clear_recorded(inm);
1021 			timer = min(inm->in6m_timer, timer);
1022 		}
1023 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1024 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1025 		V_current_state_timers_running6 = 1;
1026 		return (retval);
1027 	}
1028 
1029 	/*
1030 	 * Deal with the case where a group-and-source-specific query has
1031 	 * been received but a group-specific query is already pending.
1032 	 */
1033 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1034 		timer = min(inm->in6m_timer, timer);
1035 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1036 		V_current_state_timers_running6 = 1;
1037 		return (retval);
1038 	}
1039 
1040 	/*
1041 	 * Finally, deal with the case where a group-and-source-specific
1042 	 * query has been received, where a response to a previous g-s-r
1043 	 * query exists, or none exists.
1044 	 * In this case, we need to parse the source-list which the Querier
1045 	 * has provided us with and check if we have any source list filter
1046 	 * entries at T1 for these sources. If we do not, there is no need
1047 	 * schedule a report and the query may be dropped.
1048 	 * If we do, we must record them and schedule a current-state
1049 	 * report for those sources.
1050 	 */
1051 	if (inm->in6m_nsrc > 0) {
1052 		struct mbuf		*m;
1053 		uint8_t			*sp;
1054 		int			 i, nrecorded;
1055 		int			 soff;
1056 
1057 		m = m0;
1058 		soff = off + sizeof(struct mldv2_query);
1059 		nrecorded = 0;
1060 		for (i = 0; i < nsrc; i++) {
1061 			sp = mtod(m, uint8_t *) + soff;
1062 			retval = in6m_record_source(inm,
1063 			    (const struct in6_addr *)sp);
1064 			if (retval < 0)
1065 				break;
1066 			nrecorded += retval;
1067 			soff += sizeof(struct in6_addr);
1068 			if (soff >= m->m_len) {
1069 				soff = soff - m->m_len;
1070 				m = m->m_next;
1071 				if (m == NULL)
1072 					break;
1073 			}
1074 		}
1075 		if (nrecorded > 0) {
1076 			CTR1(KTR_MLD,
1077 			    "%s: schedule response to SG query", __func__);
1078 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1079 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1080 			V_current_state_timers_running6 = 1;
1081 		}
1082 	}
1083 
1084 	return (retval);
1085 }
1086 
1087 /*
1088  * Process a received MLDv1 host membership report.
1089  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1090  *
1091  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1092  * mld_addr. This is OK as we own the mbuf chain.
1093  */
1094 static int
1095 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1096     /*const*/ struct mld_hdr *mld)
1097 {
1098 	struct in6_addr		 src, dst;
1099 	struct in6_ifaddr	*ia;
1100 	struct in6_multi	*inm;
1101 #ifdef KTR
1102 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1103 #endif
1104 
1105 	if (!mld_v1enable) {
1106 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1107 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1108 		    ifp, if_name(ifp));
1109 		return (0);
1110 	}
1111 
1112 	if (ifp->if_flags & IFF_LOOPBACK)
1113 		return (0);
1114 
1115 	/*
1116 	 * MLDv1 reports must originate from a host's link-local address,
1117 	 * or the unspecified address (when booting).
1118 	 */
1119 	src = ip6->ip6_src;
1120 	in6_clearscope(&src);
1121 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1122 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1123 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1124 		    ifp, if_name(ifp));
1125 		return (EINVAL);
1126 	}
1127 
1128 	/*
1129 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1130 	 * group, and must be directed to the group itself.
1131 	 */
1132 	dst = ip6->ip6_dst;
1133 	in6_clearscope(&dst);
1134 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1135 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1136 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1137 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1138 		    ifp, if_name(ifp));
1139 		return (EINVAL);
1140 	}
1141 
1142 	/*
1143 	 * Make sure we don't hear our own membership report, as fast
1144 	 * leave requires knowing that we are the only member of a
1145 	 * group. Assume we used the link-local address if available,
1146 	 * otherwise look for ::.
1147 	 *
1148 	 * XXX Note that scope ID comparison is needed for the address
1149 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1150 	 * performed for the on-wire address.
1151 	 */
1152 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1153 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1154 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1155 		if (ia != NULL)
1156 			ifa_free(&ia->ia_ifa);
1157 		return (0);
1158 	}
1159 	if (ia != NULL)
1160 		ifa_free(&ia->ia_ifa);
1161 
1162 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1163 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1164 
1165 	/*
1166 	 * Embed scope ID of receiving interface in MLD query for lookup
1167 	 * whilst we don't hold other locks (due to KAME locking lameness).
1168 	 */
1169 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1170 		in6_setscope(&mld->mld_addr, ifp, NULL);
1171 
1172 	IN6_MULTI_LIST_LOCK();
1173 	MLD_LOCK();
1174 	IF_ADDR_RLOCK(ifp);
1175 
1176 	/*
1177 	 * MLDv1 report suppression.
1178 	 * If we are a member of this group, and our membership should be
1179 	 * reported, and our group timer is pending or about to be reset,
1180 	 * stop our group timer by transitioning to the 'lazy' state.
1181 	 */
1182 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1183 	if (inm != NULL) {
1184 		struct mld_ifsoftc *mli;
1185 
1186 		mli = inm->in6m_mli;
1187 		KASSERT(mli != NULL,
1188 		    ("%s: no mli for ifp %p", __func__, ifp));
1189 
1190 		/*
1191 		 * If we are in MLDv2 host mode, do not allow the
1192 		 * other host's MLDv1 report to suppress our reports.
1193 		 */
1194 		if (mli->mli_version == MLD_VERSION_2)
1195 			goto out_locked;
1196 
1197 		inm->in6m_timer = 0;
1198 
1199 		switch (inm->in6m_state) {
1200 		case MLD_NOT_MEMBER:
1201 		case MLD_SILENT_MEMBER:
1202 		case MLD_SLEEPING_MEMBER:
1203 			break;
1204 		case MLD_REPORTING_MEMBER:
1205 		case MLD_IDLE_MEMBER:
1206 		case MLD_AWAKENING_MEMBER:
1207 			CTR3(KTR_MLD,
1208 			    "report suppressed for %s on ifp %p(%s)",
1209 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1210 			    ifp, if_name(ifp));
1211 		case MLD_LAZY_MEMBER:
1212 			inm->in6m_state = MLD_LAZY_MEMBER;
1213 			break;
1214 		case MLD_G_QUERY_PENDING_MEMBER:
1215 		case MLD_SG_QUERY_PENDING_MEMBER:
1216 		case MLD_LEAVING_MEMBER:
1217 			break;
1218 		}
1219 	}
1220 
1221 out_locked:
1222 	IF_ADDR_RUNLOCK(ifp);
1223 	MLD_UNLOCK();
1224 	IN6_MULTI_LIST_UNLOCK();
1225 
1226 	/* XXX Clear embedded scope ID as userland won't expect it. */
1227 	in6_clearscope(&mld->mld_addr);
1228 
1229 	return (0);
1230 }
1231 
1232 /*
1233  * MLD input path.
1234  *
1235  * Assume query messages which fit in a single ICMPv6 message header
1236  * have been pulled up.
1237  * Assume that userland will want to see the message, even if it
1238  * otherwise fails kernel input validation; do not free it.
1239  * Pullup may however free the mbuf chain m if it fails.
1240  *
1241  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1242  */
1243 int
1244 mld_input(struct mbuf *m, int off, int icmp6len)
1245 {
1246 	struct ifnet	*ifp;
1247 	struct ip6_hdr	*ip6;
1248 	struct mld_hdr	*mld;
1249 	int		 mldlen;
1250 
1251 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1252 
1253 	ifp = m->m_pkthdr.rcvif;
1254 
1255 	ip6 = mtod(m, struct ip6_hdr *);
1256 
1257 	/* Pullup to appropriate size. */
1258 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1259 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1260 	    icmp6len >= sizeof(struct mldv2_query)) {
1261 		mldlen = sizeof(struct mldv2_query);
1262 	} else {
1263 		mldlen = sizeof(struct mld_hdr);
1264 	}
1265 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1266 	if (mld == NULL) {
1267 		ICMP6STAT_INC(icp6s_badlen);
1268 		return (IPPROTO_DONE);
1269 	}
1270 
1271 	/*
1272 	 * Userland needs to see all of this traffic for implementing
1273 	 * the endpoint discovery portion of multicast routing.
1274 	 */
1275 	switch (mld->mld_type) {
1276 	case MLD_LISTENER_QUERY:
1277 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1278 		if (icmp6len == sizeof(struct mld_hdr)) {
1279 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1280 				return (0);
1281 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1282 			if (mld_v2_input_query(ifp, ip6, m, off,
1283 			    icmp6len) != 0)
1284 				return (0);
1285 		}
1286 		break;
1287 	case MLD_LISTENER_REPORT:
1288 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1289 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1290 			return (0);
1291 		break;
1292 	case MLDV2_LISTENER_REPORT:
1293 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1294 		break;
1295 	case MLD_LISTENER_DONE:
1296 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1297 		break;
1298 	default:
1299 		break;
1300 	}
1301 
1302 	return (0);
1303 }
1304 
1305 /*
1306  * Fast timeout handler (global).
1307  * VIMAGE: Timeout handlers are expected to service all vimages.
1308  */
1309 void
1310 mld_fasttimo(void)
1311 {
1312 	VNET_ITERATOR_DECL(vnet_iter);
1313 
1314 	VNET_LIST_RLOCK_NOSLEEP();
1315 	VNET_FOREACH(vnet_iter) {
1316 		CURVNET_SET(vnet_iter);
1317 		mld_fasttimo_vnet();
1318 		CURVNET_RESTORE();
1319 	}
1320 	VNET_LIST_RUNLOCK_NOSLEEP();
1321 }
1322 
1323 /*
1324  * Fast timeout handler (per-vnet).
1325  *
1326  * VIMAGE: Assume caller has set up our curvnet.
1327  */
1328 static void
1329 mld_fasttimo_vnet(void)
1330 {
1331 	struct mbufq		 scq;	/* State-change packets */
1332 	struct mbufq		 qrq;	/* Query response packets */
1333 	struct ifnet		*ifp;
1334 	struct mld_ifsoftc	*mli;
1335 	struct ifmultiaddr	*ifma, *next;
1336 	struct in6_multi	*inm, *tinm;
1337 	struct in6_multi_head inmh;
1338 	int			 uri_fasthz;
1339 
1340 	uri_fasthz = 0;
1341 
1342 	/*
1343 	 * Quick check to see if any work needs to be done, in order to
1344 	 * minimize the overhead of fasttimo processing.
1345 	 * SMPng: XXX Unlocked reads.
1346 	 */
1347 	if (!V_current_state_timers_running6 &&
1348 	    !V_interface_timers_running6 &&
1349 	    !V_state_change_timers_running6)
1350 		return;
1351 
1352 	SLIST_INIT(&inmh);
1353 	IN6_MULTI_LIST_LOCK();
1354 	MLD_LOCK();
1355 
1356 	/*
1357 	 * MLDv2 General Query response timer processing.
1358 	 */
1359 	if (V_interface_timers_running6) {
1360 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1361 
1362 		V_interface_timers_running6 = 0;
1363 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1364 			if (mli->mli_v2_timer == 0) {
1365 				/* Do nothing. */
1366 			} else if (--mli->mli_v2_timer == 0) {
1367 				mld_v2_dispatch_general_query(mli);
1368 			} else {
1369 				V_interface_timers_running6 = 1;
1370 			}
1371 		}
1372 	}
1373 
1374 	if (!V_current_state_timers_running6 &&
1375 	    !V_state_change_timers_running6)
1376 		goto out_locked;
1377 
1378 	V_current_state_timers_running6 = 0;
1379 	V_state_change_timers_running6 = 0;
1380 
1381 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1382 
1383 	/*
1384 	 * MLD host report and state-change timer processing.
1385 	 * Note: Processing a v2 group timer may remove a node.
1386 	 */
1387 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1388 		ifp = mli->mli_ifp;
1389 
1390 		if (mli->mli_version == MLD_VERSION_2) {
1391 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1392 			    PR_FASTHZ);
1393 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1394 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1395 		}
1396 
1397 		IF_ADDR_WLOCK(ifp);
1398 	restart:
1399 		CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1400 			if (ifma->ifma_addr->sa_family != AF_INET6 ||
1401 			    ifma->ifma_protospec == NULL)
1402 				continue;
1403 			inm = (struct in6_multi *)ifma->ifma_protospec;
1404 			switch (mli->mli_version) {
1405 			case MLD_VERSION_1:
1406 				mld_v1_process_group_timer(&inmh, inm);
1407 				break;
1408 			case MLD_VERSION_2:
1409 				mld_v2_process_group_timers(&inmh, &qrq,
1410 				    &scq, inm, uri_fasthz);
1411 				break;
1412 			}
1413 			if (__predict_false(ifma6_restart)) {
1414 				ifma6_restart = false;
1415 				goto restart;
1416 			}
1417 		}
1418 		IF_ADDR_WUNLOCK(ifp);
1419 
1420 		switch (mli->mli_version) {
1421 		case MLD_VERSION_1:
1422 			/*
1423 			 * Transmit reports for this lifecycle.  This
1424 			 * is done while not holding IF_ADDR_LOCK
1425 			 * since this can call
1426 			 * in6ifa_ifpforlinklocal() which locks
1427 			 * IF_ADDR_LOCK internally as well as
1428 			 * ip6_output() to transmit a packet.
1429 			 */
1430 			SLIST_FOREACH_SAFE(inm, &inmh, in6m_nrele, tinm) {
1431 				SLIST_REMOVE_HEAD(&inmh,
1432 				    in6m_nrele);
1433 				(void)mld_v1_transmit_report(inm,
1434 				    MLD_LISTENER_REPORT);
1435 			}
1436 			break;
1437 		case MLD_VERSION_2:
1438 			mld_dispatch_queue(&qrq, 0);
1439 			mld_dispatch_queue(&scq, 0);
1440 
1441 			/*
1442 			 * Free the in_multi reference(s) for
1443 			 * this lifecycle.
1444 			 */
1445 			in6m_release_list_deferred(&inmh);
1446 			break;
1447 		}
1448 	}
1449 
1450 out_locked:
1451 	MLD_UNLOCK();
1452 	IN6_MULTI_LIST_UNLOCK();
1453 }
1454 
1455 /*
1456  * Update host report group timer.
1457  * Will update the global pending timer flags.
1458  */
1459 static void
1460 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1461 {
1462 	int report_timer_expired;
1463 
1464 	IN6_MULTI_LIST_LOCK_ASSERT();
1465 	MLD_LOCK_ASSERT();
1466 
1467 	if (inm->in6m_timer == 0) {
1468 		report_timer_expired = 0;
1469 	} else if (--inm->in6m_timer == 0) {
1470 		report_timer_expired = 1;
1471 	} else {
1472 		V_current_state_timers_running6 = 1;
1473 		return;
1474 	}
1475 
1476 	switch (inm->in6m_state) {
1477 	case MLD_NOT_MEMBER:
1478 	case MLD_SILENT_MEMBER:
1479 	case MLD_IDLE_MEMBER:
1480 	case MLD_LAZY_MEMBER:
1481 	case MLD_SLEEPING_MEMBER:
1482 	case MLD_AWAKENING_MEMBER:
1483 		break;
1484 	case MLD_REPORTING_MEMBER:
1485 		if (report_timer_expired) {
1486 			inm->in6m_state = MLD_IDLE_MEMBER;
1487 			in6m_disconnect(inm);
1488 			in6m_rele_locked(inmh, inm);
1489 		}
1490 		break;
1491 	case MLD_G_QUERY_PENDING_MEMBER:
1492 	case MLD_SG_QUERY_PENDING_MEMBER:
1493 	case MLD_LEAVING_MEMBER:
1494 		break;
1495 	}
1496 }
1497 
1498 /*
1499  * Update a group's timers for MLDv2.
1500  * Will update the global pending timer flags.
1501  * Note: Unlocked read from mli.
1502  */
1503 static void
1504 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1505     struct mbufq *qrq, struct mbufq *scq,
1506     struct in6_multi *inm, const int uri_fasthz)
1507 {
1508 	int query_response_timer_expired;
1509 	int state_change_retransmit_timer_expired;
1510 #ifdef KTR
1511 	char ip6tbuf[INET6_ADDRSTRLEN];
1512 #endif
1513 
1514 	IN6_MULTI_LIST_LOCK_ASSERT();
1515 	MLD_LOCK_ASSERT();
1516 
1517 	query_response_timer_expired = 0;
1518 	state_change_retransmit_timer_expired = 0;
1519 
1520 	/*
1521 	 * During a transition from compatibility mode back to MLDv2,
1522 	 * a group record in REPORTING state may still have its group
1523 	 * timer active. This is a no-op in this function; it is easier
1524 	 * to deal with it here than to complicate the slow-timeout path.
1525 	 */
1526 	if (inm->in6m_timer == 0) {
1527 		query_response_timer_expired = 0;
1528 	} else if (--inm->in6m_timer == 0) {
1529 		query_response_timer_expired = 1;
1530 	} else {
1531 		V_current_state_timers_running6 = 1;
1532 	}
1533 
1534 	if (inm->in6m_sctimer == 0) {
1535 		state_change_retransmit_timer_expired = 0;
1536 	} else if (--inm->in6m_sctimer == 0) {
1537 		state_change_retransmit_timer_expired = 1;
1538 	} else {
1539 		V_state_change_timers_running6 = 1;
1540 	}
1541 
1542 	/* We are in fasttimo, so be quick about it. */
1543 	if (!state_change_retransmit_timer_expired &&
1544 	    !query_response_timer_expired)
1545 		return;
1546 
1547 	switch (inm->in6m_state) {
1548 	case MLD_NOT_MEMBER:
1549 	case MLD_SILENT_MEMBER:
1550 	case MLD_SLEEPING_MEMBER:
1551 	case MLD_LAZY_MEMBER:
1552 	case MLD_AWAKENING_MEMBER:
1553 	case MLD_IDLE_MEMBER:
1554 		break;
1555 	case MLD_G_QUERY_PENDING_MEMBER:
1556 	case MLD_SG_QUERY_PENDING_MEMBER:
1557 		/*
1558 		 * Respond to a previously pending Group-Specific
1559 		 * or Group-and-Source-Specific query by enqueueing
1560 		 * the appropriate Current-State report for
1561 		 * immediate transmission.
1562 		 */
1563 		if (query_response_timer_expired) {
1564 			int retval;
1565 
1566 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1567 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1568 			    0);
1569 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1570 			    __func__, retval);
1571 			inm->in6m_state = MLD_REPORTING_MEMBER;
1572 			in6m_clear_recorded(inm);
1573 		}
1574 		/* FALLTHROUGH */
1575 	case MLD_REPORTING_MEMBER:
1576 	case MLD_LEAVING_MEMBER:
1577 		if (state_change_retransmit_timer_expired) {
1578 			/*
1579 			 * State-change retransmission timer fired.
1580 			 * If there are any further pending retransmissions,
1581 			 * set the global pending state-change flag, and
1582 			 * reset the timer.
1583 			 */
1584 			if (--inm->in6m_scrv > 0) {
1585 				inm->in6m_sctimer = uri_fasthz;
1586 				V_state_change_timers_running6 = 1;
1587 			}
1588 			/*
1589 			 * Retransmit the previously computed state-change
1590 			 * report. If there are no further pending
1591 			 * retransmissions, the mbuf queue will be consumed.
1592 			 * Update T0 state to T1 as we have now sent
1593 			 * a state-change.
1594 			 */
1595 			(void)mld_v2_merge_state_changes(inm, scq);
1596 
1597 			in6m_commit(inm);
1598 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1599 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1600 			    if_name(inm->in6m_ifp));
1601 
1602 			/*
1603 			 * If we are leaving the group for good, make sure
1604 			 * we release MLD's reference to it.
1605 			 * This release must be deferred using a SLIST,
1606 			 * as we are called from a loop which traverses
1607 			 * the in_ifmultiaddr TAILQ.
1608 			 */
1609 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1610 			    inm->in6m_scrv == 0) {
1611 				inm->in6m_state = MLD_NOT_MEMBER;
1612 				in6m_disconnect(inm);
1613 				in6m_rele_locked(inmh, inm);
1614 			}
1615 		}
1616 		break;
1617 	}
1618 }
1619 
1620 /*
1621  * Switch to a different version on the given interface,
1622  * as per Section 9.12.
1623  */
1624 static void
1625 mld_set_version(struct mld_ifsoftc *mli, const int version)
1626 {
1627 	int old_version_timer;
1628 
1629 	MLD_LOCK_ASSERT();
1630 
1631 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1632 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1633 
1634 	if (version == MLD_VERSION_1) {
1635 		/*
1636 		 * Compute the "Older Version Querier Present" timer as per
1637 		 * Section 9.12.
1638 		 */
1639 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1640 		old_version_timer *= PR_SLOWHZ;
1641 		mli->mli_v1_timer = old_version_timer;
1642 	}
1643 
1644 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1645 		mli->mli_version = MLD_VERSION_1;
1646 		mld_v2_cancel_link_timers(mli);
1647 	}
1648 }
1649 
1650 /*
1651  * Cancel pending MLDv2 timers for the given link and all groups
1652  * joined on it; state-change, general-query, and group-query timers.
1653  */
1654 static void
1655 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1656 {
1657 	struct ifmultiaddr	*ifma, *next;
1658 	struct ifnet		*ifp;
1659 	struct in6_multi	*inm;
1660 	struct in6_multi_head inmh;
1661 
1662 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1663 	    mli->mli_ifp, if_name(mli->mli_ifp));
1664 
1665 	SLIST_INIT(&inmh);
1666 	IN6_MULTI_LIST_LOCK_ASSERT();
1667 	MLD_LOCK_ASSERT();
1668 
1669 	/*
1670 	 * Fast-track this potentially expensive operation
1671 	 * by checking all the global 'timer pending' flags.
1672 	 */
1673 	if (!V_interface_timers_running6 &&
1674 	    !V_state_change_timers_running6 &&
1675 	    !V_current_state_timers_running6)
1676 		return;
1677 
1678 	mli->mli_v2_timer = 0;
1679 
1680 	ifp = mli->mli_ifp;
1681 
1682 	IF_ADDR_WLOCK(ifp);
1683  restart:
1684 	CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1685 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
1686 		    ifma->ifma_protospec == NULL)
1687 			continue;
1688 		inm = (struct in6_multi *)ifma->ifma_protospec;
1689 		switch (inm->in6m_state) {
1690 		case MLD_NOT_MEMBER:
1691 		case MLD_SILENT_MEMBER:
1692 		case MLD_IDLE_MEMBER:
1693 		case MLD_LAZY_MEMBER:
1694 		case MLD_SLEEPING_MEMBER:
1695 		case MLD_AWAKENING_MEMBER:
1696 			break;
1697 		case MLD_LEAVING_MEMBER:
1698 			/*
1699 			 * If we are leaving the group and switching
1700 			 * version, we need to release the final
1701 			 * reference held for issuing the INCLUDE {}.
1702 			 */
1703 			in6m_disconnect(inm);
1704 			in6m_rele_locked(&inmh, inm);
1705 			ifma->ifma_protospec = NULL;
1706 			/* FALLTHROUGH */
1707 		case MLD_G_QUERY_PENDING_MEMBER:
1708 		case MLD_SG_QUERY_PENDING_MEMBER:
1709 			in6m_clear_recorded(inm);
1710 			/* FALLTHROUGH */
1711 		case MLD_REPORTING_MEMBER:
1712 			inm->in6m_sctimer = 0;
1713 			inm->in6m_timer = 0;
1714 			inm->in6m_state = MLD_REPORTING_MEMBER;
1715 			/*
1716 			 * Free any pending MLDv2 state-change records.
1717 			 */
1718 			mbufq_drain(&inm->in6m_scq);
1719 			break;
1720 		}
1721 		if (__predict_false(ifma6_restart)) {
1722 			ifma6_restart = false;
1723 			goto restart;
1724 		}
1725 	}
1726 	IF_ADDR_WUNLOCK(ifp);
1727 	in6m_release_list_deferred(&inmh);
1728 }
1729 
1730 /*
1731  * Global slowtimo handler.
1732  * VIMAGE: Timeout handlers are expected to service all vimages.
1733  */
1734 void
1735 mld_slowtimo(void)
1736 {
1737 	VNET_ITERATOR_DECL(vnet_iter);
1738 
1739 	VNET_LIST_RLOCK_NOSLEEP();
1740 	VNET_FOREACH(vnet_iter) {
1741 		CURVNET_SET(vnet_iter);
1742 		mld_slowtimo_vnet();
1743 		CURVNET_RESTORE();
1744 	}
1745 	VNET_LIST_RUNLOCK_NOSLEEP();
1746 }
1747 
1748 /*
1749  * Per-vnet slowtimo handler.
1750  */
1751 static void
1752 mld_slowtimo_vnet(void)
1753 {
1754 	struct mld_ifsoftc *mli;
1755 
1756 	MLD_LOCK();
1757 
1758 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1759 		mld_v1_process_querier_timers(mli);
1760 	}
1761 
1762 	MLD_UNLOCK();
1763 }
1764 
1765 /*
1766  * Update the Older Version Querier Present timers for a link.
1767  * See Section 9.12 of RFC 3810.
1768  */
1769 static void
1770 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1771 {
1772 
1773 	MLD_LOCK_ASSERT();
1774 
1775 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1776 		/*
1777 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1778 		 */
1779 		CTR5(KTR_MLD,
1780 		    "%s: transition from v%d -> v%d on %p(%s)",
1781 		    __func__, mli->mli_version, MLD_VERSION_2,
1782 		    mli->mli_ifp, if_name(mli->mli_ifp));
1783 		mli->mli_version = MLD_VERSION_2;
1784 	}
1785 }
1786 
1787 /*
1788  * Transmit an MLDv1 report immediately.
1789  */
1790 static int
1791 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1792 {
1793 	struct ifnet		*ifp;
1794 	struct in6_ifaddr	*ia;
1795 	struct ip6_hdr		*ip6;
1796 	struct mbuf		*mh, *md;
1797 	struct mld_hdr		*mld;
1798 
1799 	IN6_MULTI_LIST_LOCK_ASSERT();
1800 	MLD_LOCK_ASSERT();
1801 
1802 	ifp = in6m->in6m_ifp;
1803 	/* in process of being freed */
1804 	if (ifp == NULL)
1805 		return (0);
1806 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1807 	/* ia may be NULL if link-local address is tentative. */
1808 
1809 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1810 	if (mh == NULL) {
1811 		if (ia != NULL)
1812 			ifa_free(&ia->ia_ifa);
1813 		return (ENOMEM);
1814 	}
1815 	md = m_get(M_NOWAIT, MT_DATA);
1816 	if (md == NULL) {
1817 		m_free(mh);
1818 		if (ia != NULL)
1819 			ifa_free(&ia->ia_ifa);
1820 		return (ENOMEM);
1821 	}
1822 	mh->m_next = md;
1823 
1824 	/*
1825 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1826 	 * that ether_output() does not need to allocate another mbuf
1827 	 * for the header in the most common case.
1828 	 */
1829 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1830 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1831 	mh->m_len = sizeof(struct ip6_hdr);
1832 
1833 	ip6 = mtod(mh, struct ip6_hdr *);
1834 	ip6->ip6_flow = 0;
1835 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1836 	ip6->ip6_vfc |= IPV6_VERSION;
1837 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1838 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1839 	ip6->ip6_dst = in6m->in6m_addr;
1840 
1841 	md->m_len = sizeof(struct mld_hdr);
1842 	mld = mtod(md, struct mld_hdr *);
1843 	mld->mld_type = type;
1844 	mld->mld_code = 0;
1845 	mld->mld_cksum = 0;
1846 	mld->mld_maxdelay = 0;
1847 	mld->mld_reserved = 0;
1848 	mld->mld_addr = in6m->in6m_addr;
1849 	in6_clearscope(&mld->mld_addr);
1850 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1851 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1852 
1853 	mld_save_context(mh, ifp);
1854 	mh->m_flags |= M_MLDV1;
1855 
1856 	mld_dispatch_packet(mh);
1857 
1858 	if (ia != NULL)
1859 		ifa_free(&ia->ia_ifa);
1860 	return (0);
1861 }
1862 
1863 /*
1864  * Process a state change from the upper layer for the given IPv6 group.
1865  *
1866  * Each socket holds a reference on the in_multi in its own ip_moptions.
1867  * The socket layer will have made the necessary updates to.the group
1868  * state, it is now up to MLD to issue a state change report if there
1869  * has been any change between T0 (when the last state-change was issued)
1870  * and T1 (now).
1871  *
1872  * We use the MLDv2 state machine at group level. The MLd module
1873  * however makes the decision as to which MLD protocol version to speak.
1874  * A state change *from* INCLUDE {} always means an initial join.
1875  * A state change *to* INCLUDE {} always means a final leave.
1876  *
1877  * If delay is non-zero, and the state change is an initial multicast
1878  * join, the state change report will be delayed by 'delay' ticks
1879  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1880  * the initial MLDv2 state change report will be delayed by whichever
1881  * is sooner, a pending state-change timer or delay itself.
1882  *
1883  * VIMAGE: curvnet should have been set by caller, as this routine
1884  * is called from the socket option handlers.
1885  */
1886 int
1887 mld_change_state(struct in6_multi *inm, const int delay)
1888 {
1889 	struct mld_ifsoftc *mli;
1890 	struct ifnet *ifp;
1891 	int error;
1892 
1893 	IN6_MULTI_LIST_LOCK_ASSERT();
1894 
1895 	error = 0;
1896 
1897 	/*
1898 	 * Try to detect if the upper layer just asked us to change state
1899 	 * for an interface which has now gone away.
1900 	 */
1901 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1902 	ifp = inm->in6m_ifma->ifma_ifp;
1903 	if (ifp == NULL)
1904 		return (0);
1905 	/*
1906 	 * Sanity check that netinet6's notion of ifp is the
1907 	 * same as net's.
1908 	 */
1909 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1910 
1911 	MLD_LOCK();
1912 	mli = MLD_IFINFO(ifp);
1913 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1914 
1915 	/*
1916 	 * If we detect a state transition to or from MCAST_UNDEFINED
1917 	 * for this group, then we are starting or finishing an MLD
1918 	 * life cycle for this group.
1919 	 */
1920 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1921 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1922 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1923 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1924 			CTR1(KTR_MLD, "%s: initial join", __func__);
1925 			error = mld_initial_join(inm, mli, delay);
1926 			goto out_locked;
1927 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1928 			CTR1(KTR_MLD, "%s: final leave", __func__);
1929 			mld_final_leave(inm, mli);
1930 			goto out_locked;
1931 		}
1932 	} else {
1933 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1934 	}
1935 
1936 	error = mld_handle_state_change(inm, mli);
1937 
1938 out_locked:
1939 	MLD_UNLOCK();
1940 	return (error);
1941 }
1942 
1943 /*
1944  * Perform the initial join for an MLD group.
1945  *
1946  * When joining a group:
1947  *  If the group should have its MLD traffic suppressed, do nothing.
1948  *  MLDv1 starts sending MLDv1 host membership reports.
1949  *  MLDv2 will schedule an MLDv2 state-change report containing the
1950  *  initial state of the membership.
1951  *
1952  * If the delay argument is non-zero, then we must delay sending the
1953  * initial state change for delay ticks (in units of PR_FASTHZ).
1954  */
1955 static int
1956 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1957     const int delay)
1958 {
1959 	struct ifnet		*ifp;
1960 	struct mbufq		*mq;
1961 	int			 error, retval, syncstates;
1962 	int			 odelay;
1963 #ifdef KTR
1964 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1965 #endif
1966 
1967 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1968 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1969 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1970 
1971 	error = 0;
1972 	syncstates = 1;
1973 
1974 	ifp = inm->in6m_ifp;
1975 
1976 	IN6_MULTI_LIST_LOCK_ASSERT();
1977 	MLD_LOCK_ASSERT();
1978 
1979 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1980 
1981 	/*
1982 	 * Groups joined on loopback or marked as 'not reported',
1983 	 * enter the MLD_SILENT_MEMBER state and
1984 	 * are never reported in any protocol exchanges.
1985 	 * All other groups enter the appropriate state machine
1986 	 * for the version in use on this link.
1987 	 * A link marked as MLIF_SILENT causes MLD to be completely
1988 	 * disabled for the link.
1989 	 */
1990 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1991 	    (mli->mli_flags & MLIF_SILENT) ||
1992 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1993 		CTR1(KTR_MLD,
1994 "%s: not kicking state machine for silent group", __func__);
1995 		inm->in6m_state = MLD_SILENT_MEMBER;
1996 		inm->in6m_timer = 0;
1997 	} else {
1998 		/*
1999 		 * Deal with overlapping in_multi lifecycle.
2000 		 * If this group was LEAVING, then make sure
2001 		 * we drop the reference we picked up to keep the
2002 		 * group around for the final INCLUDE {} enqueue.
2003 		 */
2004 		if (mli->mli_version == MLD_VERSION_2 &&
2005 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2006 			inm->in6m_refcount--;
2007 		}
2008 		inm->in6m_state = MLD_REPORTING_MEMBER;
2009 
2010 		switch (mli->mli_version) {
2011 		case MLD_VERSION_1:
2012 			/*
2013 			 * If a delay was provided, only use it if
2014 			 * it is greater than the delay normally
2015 			 * used for an MLDv1 state change report,
2016 			 * and delay sending the initial MLDv1 report
2017 			 * by not transitioning to the IDLE state.
2018 			 */
2019 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2020 			if (delay) {
2021 				inm->in6m_timer = max(delay, odelay);
2022 				V_current_state_timers_running6 = 1;
2023 			} else {
2024 				inm->in6m_state = MLD_IDLE_MEMBER;
2025 				error = mld_v1_transmit_report(inm,
2026 				     MLD_LISTENER_REPORT);
2027 				if (error == 0) {
2028 					inm->in6m_timer = odelay;
2029 					V_current_state_timers_running6 = 1;
2030 				}
2031 			}
2032 			break;
2033 
2034 		case MLD_VERSION_2:
2035 			/*
2036 			 * Defer update of T0 to T1, until the first copy
2037 			 * of the state change has been transmitted.
2038 			 */
2039 			syncstates = 0;
2040 
2041 			/*
2042 			 * Immediately enqueue a State-Change Report for
2043 			 * this interface, freeing any previous reports.
2044 			 * Don't kick the timers if there is nothing to do,
2045 			 * or if an error occurred.
2046 			 */
2047 			mq = &inm->in6m_scq;
2048 			mbufq_drain(mq);
2049 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2050 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2051 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2052 			    __func__, retval);
2053 			if (retval <= 0) {
2054 				error = retval * -1;
2055 				break;
2056 			}
2057 
2058 			/*
2059 			 * Schedule transmission of pending state-change
2060 			 * report up to RV times for this link. The timer
2061 			 * will fire at the next mld_fasttimo (~200ms),
2062 			 * giving us an opportunity to merge the reports.
2063 			 *
2064 			 * If a delay was provided to this function, only
2065 			 * use this delay if sooner than the existing one.
2066 			 */
2067 			KASSERT(mli->mli_rv > 1,
2068 			   ("%s: invalid robustness %d", __func__,
2069 			    mli->mli_rv));
2070 			inm->in6m_scrv = mli->mli_rv;
2071 			if (delay) {
2072 				if (inm->in6m_sctimer > 1) {
2073 					inm->in6m_sctimer =
2074 					    min(inm->in6m_sctimer, delay);
2075 				} else
2076 					inm->in6m_sctimer = delay;
2077 			} else
2078 				inm->in6m_sctimer = 1;
2079 			V_state_change_timers_running6 = 1;
2080 
2081 			error = 0;
2082 			break;
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Only update the T0 state if state change is atomic,
2088 	 * i.e. we don't need to wait for a timer to fire before we
2089 	 * can consider the state change to have been communicated.
2090 	 */
2091 	if (syncstates) {
2092 		in6m_commit(inm);
2093 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2094 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2095 		    if_name(inm->in6m_ifp));
2096 	}
2097 
2098 	return (error);
2099 }
2100 
2101 /*
2102  * Issue an intermediate state change during the life-cycle.
2103  */
2104 static int
2105 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2106 {
2107 	struct ifnet		*ifp;
2108 	int			 retval;
2109 #ifdef KTR
2110 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2111 #endif
2112 
2113 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2114 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2115 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2116 
2117 	ifp = inm->in6m_ifp;
2118 
2119 	IN6_MULTI_LIST_LOCK_ASSERT();
2120 	MLD_LOCK_ASSERT();
2121 
2122 	KASSERT(mli && mli->mli_ifp == ifp,
2123 	    ("%s: inconsistent ifp", __func__));
2124 
2125 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2126 	    (mli->mli_flags & MLIF_SILENT) ||
2127 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2128 	    (mli->mli_version != MLD_VERSION_2)) {
2129 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2130 			CTR1(KTR_MLD,
2131 "%s: not kicking state machine for silent group", __func__);
2132 		}
2133 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2134 		in6m_commit(inm);
2135 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2136 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2137 		    if_name(inm->in6m_ifp));
2138 		return (0);
2139 	}
2140 
2141 	mbufq_drain(&inm->in6m_scq);
2142 
2143 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2144 	    (mli->mli_flags & MLIF_USEALLOW));
2145 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2146 	if (retval <= 0)
2147 		return (-retval);
2148 
2149 	/*
2150 	 * If record(s) were enqueued, start the state-change
2151 	 * report timer for this group.
2152 	 */
2153 	inm->in6m_scrv = mli->mli_rv;
2154 	inm->in6m_sctimer = 1;
2155 	V_state_change_timers_running6 = 1;
2156 
2157 	return (0);
2158 }
2159 
2160 /*
2161  * Perform the final leave for a multicast address.
2162  *
2163  * When leaving a group:
2164  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2165  *  MLDv2 enqueues a state-change report containing a transition
2166  *  to INCLUDE {} for immediate transmission.
2167  */
2168 static void
2169 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2170 {
2171 	int syncstates;
2172 #ifdef KTR
2173 	char ip6tbuf[INET6_ADDRSTRLEN];
2174 #endif
2175 
2176 	syncstates = 1;
2177 
2178 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2179 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2180 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2181 
2182 	IN6_MULTI_LIST_LOCK_ASSERT();
2183 	MLD_LOCK_ASSERT();
2184 
2185 	switch (inm->in6m_state) {
2186 	case MLD_NOT_MEMBER:
2187 	case MLD_SILENT_MEMBER:
2188 	case MLD_LEAVING_MEMBER:
2189 		/* Already leaving or left; do nothing. */
2190 		CTR1(KTR_MLD,
2191 "%s: not kicking state machine for silent group", __func__);
2192 		break;
2193 	case MLD_REPORTING_MEMBER:
2194 	case MLD_IDLE_MEMBER:
2195 	case MLD_G_QUERY_PENDING_MEMBER:
2196 	case MLD_SG_QUERY_PENDING_MEMBER:
2197 		if (mli->mli_version == MLD_VERSION_1) {
2198 #ifdef INVARIANTS
2199 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2200 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2201 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2202 			     __func__);
2203 #endif
2204 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2205 			inm->in6m_state = MLD_NOT_MEMBER;
2206 			V_current_state_timers_running6 = 1;
2207 		} else if (mli->mli_version == MLD_VERSION_2) {
2208 			/*
2209 			 * Stop group timer and all pending reports.
2210 			 * Immediately enqueue a state-change report
2211 			 * TO_IN {} to be sent on the next fast timeout,
2212 			 * giving us an opportunity to merge reports.
2213 			 */
2214 			mbufq_drain(&inm->in6m_scq);
2215 			inm->in6m_timer = 0;
2216 			inm->in6m_scrv = mli->mli_rv;
2217 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2218 			    "pending retransmissions.", __func__,
2219 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2220 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2221 			if (inm->in6m_scrv == 0) {
2222 				inm->in6m_state = MLD_NOT_MEMBER;
2223 				inm->in6m_sctimer = 0;
2224 			} else {
2225 				int retval;
2226 
2227 				in6m_acquire_locked(inm);
2228 
2229 				retval = mld_v2_enqueue_group_record(
2230 				    &inm->in6m_scq, inm, 1, 0, 0,
2231 				    (mli->mli_flags & MLIF_USEALLOW));
2232 				KASSERT(retval != 0,
2233 				    ("%s: enqueue record = %d", __func__,
2234 				     retval));
2235 
2236 				inm->in6m_state = MLD_LEAVING_MEMBER;
2237 				inm->in6m_sctimer = 1;
2238 				V_state_change_timers_running6 = 1;
2239 				syncstates = 0;
2240 			}
2241 			break;
2242 		}
2243 		break;
2244 	case MLD_LAZY_MEMBER:
2245 	case MLD_SLEEPING_MEMBER:
2246 	case MLD_AWAKENING_MEMBER:
2247 		/* Our reports are suppressed; do nothing. */
2248 		break;
2249 	}
2250 
2251 	if (syncstates) {
2252 		in6m_commit(inm);
2253 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2254 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2255 		    if_name(inm->in6m_ifp));
2256 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2257 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2258 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2259 	}
2260 }
2261 
2262 /*
2263  * Enqueue an MLDv2 group record to the given output queue.
2264  *
2265  * If is_state_change is zero, a current-state record is appended.
2266  * If is_state_change is non-zero, a state-change report is appended.
2267  *
2268  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2269  * If is_group_query is zero, and if there is a packet with free space
2270  * at the tail of the queue, it will be appended to providing there
2271  * is enough free space.
2272  * Otherwise a new mbuf packet chain is allocated.
2273  *
2274  * If is_source_query is non-zero, each source is checked to see if
2275  * it was recorded for a Group-Source query, and will be omitted if
2276  * it is not both in-mode and recorded.
2277  *
2278  * If use_block_allow is non-zero, state change reports for initial join
2279  * and final leave, on an inclusive mode group with a source list, will be
2280  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2281  *
2282  * The function will attempt to allocate leading space in the packet
2283  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2284  *
2285  * If successful the size of all data appended to the queue is returned,
2286  * otherwise an error code less than zero is returned, or zero if
2287  * no record(s) were appended.
2288  */
2289 static int
2290 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2291     const int is_state_change, const int is_group_query,
2292     const int is_source_query, const int use_block_allow)
2293 {
2294 	struct mldv2_record	 mr;
2295 	struct mldv2_record	*pmr;
2296 	struct ifnet		*ifp;
2297 	struct ip6_msource	*ims, *nims;
2298 	struct mbuf		*m0, *m, *md;
2299 	int			 is_filter_list_change;
2300 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2301 	int			 record_has_sources;
2302 	int			 now;
2303 	int			 type;
2304 	uint8_t			 mode;
2305 #ifdef KTR
2306 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2307 #endif
2308 
2309 	IN6_MULTI_LIST_LOCK_ASSERT();
2310 
2311 	ifp = inm->in6m_ifp;
2312 	is_filter_list_change = 0;
2313 	m = NULL;
2314 	m0 = NULL;
2315 	m0srcs = 0;
2316 	msrcs = 0;
2317 	nbytes = 0;
2318 	nims = NULL;
2319 	record_has_sources = 1;
2320 	pmr = NULL;
2321 	type = MLD_DO_NOTHING;
2322 	mode = inm->in6m_st[1].iss_fmode;
2323 
2324 	/*
2325 	 * If we did not transition out of ASM mode during t0->t1,
2326 	 * and there are no source nodes to process, we can skip
2327 	 * the generation of source records.
2328 	 */
2329 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2330 	    inm->in6m_nsrc == 0)
2331 		record_has_sources = 0;
2332 
2333 	if (is_state_change) {
2334 		/*
2335 		 * Queue a state change record.
2336 		 * If the mode did not change, and there are non-ASM
2337 		 * listeners or source filters present,
2338 		 * we potentially need to issue two records for the group.
2339 		 * If there are ASM listeners, and there was no filter
2340 		 * mode transition of any kind, do nothing.
2341 		 *
2342 		 * If we are transitioning to MCAST_UNDEFINED, we need
2343 		 * not send any sources. A transition to/from this state is
2344 		 * considered inclusive with some special treatment.
2345 		 *
2346 		 * If we are rewriting initial joins/leaves to use
2347 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2348 		 * we need to send sources in all cases.
2349 		 */
2350 		if (mode != inm->in6m_st[0].iss_fmode) {
2351 			if (mode == MCAST_EXCLUDE) {
2352 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2353 				    __func__);
2354 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2355 			} else {
2356 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2357 				    __func__);
2358 				if (use_block_allow) {
2359 					/*
2360 					 * XXX
2361 					 * Here we're interested in state
2362 					 * edges either direction between
2363 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2364 					 * Perhaps we should just check
2365 					 * the group state, rather than
2366 					 * the filter mode.
2367 					 */
2368 					if (mode == MCAST_UNDEFINED) {
2369 						type = MLD_BLOCK_OLD_SOURCES;
2370 					} else {
2371 						type = MLD_ALLOW_NEW_SOURCES;
2372 					}
2373 				} else {
2374 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2375 					if (mode == MCAST_UNDEFINED)
2376 						record_has_sources = 0;
2377 				}
2378 			}
2379 		} else {
2380 			if (record_has_sources) {
2381 				is_filter_list_change = 1;
2382 			} else {
2383 				type = MLD_DO_NOTHING;
2384 			}
2385 		}
2386 	} else {
2387 		/*
2388 		 * Queue a current state record.
2389 		 */
2390 		if (mode == MCAST_EXCLUDE) {
2391 			type = MLD_MODE_IS_EXCLUDE;
2392 		} else if (mode == MCAST_INCLUDE) {
2393 			type = MLD_MODE_IS_INCLUDE;
2394 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2395 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2396 			     __func__, inm, inm->in6m_st[1].iss_asm));
2397 		}
2398 	}
2399 
2400 	/*
2401 	 * Generate the filter list changes using a separate function.
2402 	 */
2403 	if (is_filter_list_change)
2404 		return (mld_v2_enqueue_filter_change(mq, inm));
2405 
2406 	if (type == MLD_DO_NOTHING) {
2407 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2408 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2409 		    if_name(inm->in6m_ifp));
2410 		return (0);
2411 	}
2412 
2413 	/*
2414 	 * If any sources are present, we must be able to fit at least
2415 	 * one in the trailing space of the tail packet's mbuf,
2416 	 * ideally more.
2417 	 */
2418 	minrec0len = sizeof(struct mldv2_record);
2419 	if (record_has_sources)
2420 		minrec0len += sizeof(struct in6_addr);
2421 
2422 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2423 	    mld_rec_type_to_str(type),
2424 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2425 	    if_name(inm->in6m_ifp));
2426 
2427 	/*
2428 	 * Check if we have a packet in the tail of the queue for this
2429 	 * group into which the first group record for this group will fit.
2430 	 * Otherwise allocate a new packet.
2431 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2432 	 * Note: Group records for G/GSR query responses MUST be sent
2433 	 * in their own packet.
2434 	 */
2435 	m0 = mbufq_last(mq);
2436 	if (!is_group_query &&
2437 	    m0 != NULL &&
2438 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2439 	    (m0->m_pkthdr.len + minrec0len) <
2440 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2441 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2442 			    sizeof(struct mldv2_record)) /
2443 			    sizeof(struct in6_addr);
2444 		m = m0;
2445 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2446 	} else {
2447 		if (mbufq_full(mq)) {
2448 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2449 			return (-ENOMEM);
2450 		}
2451 		m = NULL;
2452 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2453 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2454 		if (!is_state_change && !is_group_query)
2455 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2456 		if (m == NULL)
2457 			m = m_gethdr(M_NOWAIT, MT_DATA);
2458 		if (m == NULL)
2459 			return (-ENOMEM);
2460 
2461 		mld_save_context(m, ifp);
2462 
2463 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2464 	}
2465 
2466 	/*
2467 	 * Append group record.
2468 	 * If we have sources, we don't know how many yet.
2469 	 */
2470 	mr.mr_type = type;
2471 	mr.mr_datalen = 0;
2472 	mr.mr_numsrc = 0;
2473 	mr.mr_addr = inm->in6m_addr;
2474 	in6_clearscope(&mr.mr_addr);
2475 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2476 		if (m != m0)
2477 			m_freem(m);
2478 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2479 		return (-ENOMEM);
2480 	}
2481 	nbytes += sizeof(struct mldv2_record);
2482 
2483 	/*
2484 	 * Append as many sources as will fit in the first packet.
2485 	 * If we are appending to a new packet, the chain allocation
2486 	 * may potentially use clusters; use m_getptr() in this case.
2487 	 * If we are appending to an existing packet, we need to obtain
2488 	 * a pointer to the group record after m_append(), in case a new
2489 	 * mbuf was allocated.
2490 	 *
2491 	 * Only append sources which are in-mode at t1. If we are
2492 	 * transitioning to MCAST_UNDEFINED state on the group, and
2493 	 * use_block_allow is zero, do not include source entries.
2494 	 * Otherwise, we need to include this source in the report.
2495 	 *
2496 	 * Only report recorded sources in our filter set when responding
2497 	 * to a group-source query.
2498 	 */
2499 	if (record_has_sources) {
2500 		if (m == m0) {
2501 			md = m_last(m);
2502 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2503 			    md->m_len - nbytes);
2504 		} else {
2505 			md = m_getptr(m, 0, &off);
2506 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2507 			    off);
2508 		}
2509 		msrcs = 0;
2510 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2511 		    nims) {
2512 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2513 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2514 			now = im6s_get_mode(inm, ims, 1);
2515 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2516 			if ((now != mode) ||
2517 			    (now == mode &&
2518 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2519 				CTR1(KTR_MLD, "%s: skip node", __func__);
2520 				continue;
2521 			}
2522 			if (is_source_query && ims->im6s_stp == 0) {
2523 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2524 				    __func__);
2525 				continue;
2526 			}
2527 			CTR1(KTR_MLD, "%s: append node", __func__);
2528 			if (!m_append(m, sizeof(struct in6_addr),
2529 			    (void *)&ims->im6s_addr)) {
2530 				if (m != m0)
2531 					m_freem(m);
2532 				CTR1(KTR_MLD, "%s: m_append() failed.",
2533 				    __func__);
2534 				return (-ENOMEM);
2535 			}
2536 			nbytes += sizeof(struct in6_addr);
2537 			++msrcs;
2538 			if (msrcs == m0srcs)
2539 				break;
2540 		}
2541 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2542 		    msrcs);
2543 		pmr->mr_numsrc = htons(msrcs);
2544 		nbytes += (msrcs * sizeof(struct in6_addr));
2545 	}
2546 
2547 	if (is_source_query && msrcs == 0) {
2548 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2549 		if (m != m0)
2550 			m_freem(m);
2551 		return (0);
2552 	}
2553 
2554 	/*
2555 	 * We are good to go with first packet.
2556 	 */
2557 	if (m != m0) {
2558 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2559 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2560 		mbufq_enqueue(mq, m);
2561 	} else
2562 		m->m_pkthdr.PH_vt.vt_nrecs++;
2563 
2564 	/*
2565 	 * No further work needed if no source list in packet(s).
2566 	 */
2567 	if (!record_has_sources)
2568 		return (nbytes);
2569 
2570 	/*
2571 	 * Whilst sources remain to be announced, we need to allocate
2572 	 * a new packet and fill out as many sources as will fit.
2573 	 * Always try for a cluster first.
2574 	 */
2575 	while (nims != NULL) {
2576 		if (mbufq_full(mq)) {
2577 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2578 			return (-ENOMEM);
2579 		}
2580 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2581 		if (m == NULL)
2582 			m = m_gethdr(M_NOWAIT, MT_DATA);
2583 		if (m == NULL)
2584 			return (-ENOMEM);
2585 		mld_save_context(m, ifp);
2586 		md = m_getptr(m, 0, &off);
2587 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2588 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2589 
2590 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2591 			if (m != m0)
2592 				m_freem(m);
2593 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2594 			return (-ENOMEM);
2595 		}
2596 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2597 		nbytes += sizeof(struct mldv2_record);
2598 
2599 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2600 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2601 
2602 		msrcs = 0;
2603 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2604 			CTR2(KTR_MLD, "%s: visit node %s",
2605 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2606 			now = im6s_get_mode(inm, ims, 1);
2607 			if ((now != mode) ||
2608 			    (now == mode &&
2609 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2610 				CTR1(KTR_MLD, "%s: skip node", __func__);
2611 				continue;
2612 			}
2613 			if (is_source_query && ims->im6s_stp == 0) {
2614 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2615 				    __func__);
2616 				continue;
2617 			}
2618 			CTR1(KTR_MLD, "%s: append node", __func__);
2619 			if (!m_append(m, sizeof(struct in6_addr),
2620 			    (void *)&ims->im6s_addr)) {
2621 				if (m != m0)
2622 					m_freem(m);
2623 				CTR1(KTR_MLD, "%s: m_append() failed.",
2624 				    __func__);
2625 				return (-ENOMEM);
2626 			}
2627 			++msrcs;
2628 			if (msrcs == m0srcs)
2629 				break;
2630 		}
2631 		pmr->mr_numsrc = htons(msrcs);
2632 		nbytes += (msrcs * sizeof(struct in6_addr));
2633 
2634 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2635 		mbufq_enqueue(mq, m);
2636 	}
2637 
2638 	return (nbytes);
2639 }
2640 
2641 /*
2642  * Type used to mark record pass completion.
2643  * We exploit the fact we can cast to this easily from the
2644  * current filter modes on each ip_msource node.
2645  */
2646 typedef enum {
2647 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2648 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2649 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2650 	REC_FULL = REC_ALLOW | REC_BLOCK
2651 } rectype_t;
2652 
2653 /*
2654  * Enqueue an MLDv2 filter list change to the given output queue.
2655  *
2656  * Source list filter state is held in an RB-tree. When the filter list
2657  * for a group is changed without changing its mode, we need to compute
2658  * the deltas between T0 and T1 for each source in the filter set,
2659  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2660  *
2661  * As we may potentially queue two record types, and the entire R-B tree
2662  * needs to be walked at once, we break this out into its own function
2663  * so we can generate a tightly packed queue of packets.
2664  *
2665  * XXX This could be written to only use one tree walk, although that makes
2666  * serializing into the mbuf chains a bit harder. For now we do two walks
2667  * which makes things easier on us, and it may or may not be harder on
2668  * the L2 cache.
2669  *
2670  * If successful the size of all data appended to the queue is returned,
2671  * otherwise an error code less than zero is returned, or zero if
2672  * no record(s) were appended.
2673  */
2674 static int
2675 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2676 {
2677 	static const int MINRECLEN =
2678 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2679 	struct ifnet		*ifp;
2680 	struct mldv2_record	 mr;
2681 	struct mldv2_record	*pmr;
2682 	struct ip6_msource	*ims, *nims;
2683 	struct mbuf		*m, *m0, *md;
2684 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2685 	int			 nallow, nblock;
2686 	uint8_t			 mode, now, then;
2687 	rectype_t		 crt, drt, nrt;
2688 #ifdef KTR
2689 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2690 #endif
2691 
2692 	IN6_MULTI_LIST_LOCK_ASSERT();
2693 
2694 	if (inm->in6m_nsrc == 0 ||
2695 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2696 		return (0);
2697 
2698 	ifp = inm->in6m_ifp;			/* interface */
2699 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2700 	crt = REC_NONE;	/* current group record type */
2701 	drt = REC_NONE;	/* mask of completed group record types */
2702 	nrt = REC_NONE;	/* record type for current node */
2703 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2704 	npbytes = 0;	/* # of bytes appended this packet */
2705 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2706 	rsrcs = 0;	/* # sources encoded in current record */
2707 	schanged = 0;	/* # nodes encoded in overall filter change */
2708 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2709 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2710 	nims = NULL;	/* next tree node pointer */
2711 
2712 	/*
2713 	 * For each possible filter record mode.
2714 	 * The first kind of source we encounter tells us which
2715 	 * is the first kind of record we start appending.
2716 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2717 	 * as the inverse of the group's filter mode.
2718 	 */
2719 	while (drt != REC_FULL) {
2720 		do {
2721 			m0 = mbufq_last(mq);
2722 			if (m0 != NULL &&
2723 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2724 			     MLD_V2_REPORT_MAXRECS) &&
2725 			    (m0->m_pkthdr.len + MINRECLEN) <
2726 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2727 				m = m0;
2728 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2729 					    sizeof(struct mldv2_record)) /
2730 					    sizeof(struct in6_addr);
2731 				CTR1(KTR_MLD,
2732 				    "%s: use previous packet", __func__);
2733 			} else {
2734 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2735 				if (m == NULL)
2736 					m = m_gethdr(M_NOWAIT, MT_DATA);
2737 				if (m == NULL) {
2738 					CTR1(KTR_MLD,
2739 					    "%s: m_get*() failed", __func__);
2740 					return (-ENOMEM);
2741 				}
2742 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2743 				mld_save_context(m, ifp);
2744 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2745 				    sizeof(struct mldv2_record)) /
2746 				    sizeof(struct in6_addr);
2747 				npbytes = 0;
2748 				CTR1(KTR_MLD,
2749 				    "%s: allocated new packet", __func__);
2750 			}
2751 			/*
2752 			 * Append the MLD group record header to the
2753 			 * current packet's data area.
2754 			 * Recalculate pointer to free space for next
2755 			 * group record, in case m_append() allocated
2756 			 * a new mbuf or cluster.
2757 			 */
2758 			memset(&mr, 0, sizeof(mr));
2759 			mr.mr_addr = inm->in6m_addr;
2760 			in6_clearscope(&mr.mr_addr);
2761 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2762 				if (m != m0)
2763 					m_freem(m);
2764 				CTR1(KTR_MLD,
2765 				    "%s: m_append() failed", __func__);
2766 				return (-ENOMEM);
2767 			}
2768 			npbytes += sizeof(struct mldv2_record);
2769 			if (m != m0) {
2770 				/* new packet; offset in chain */
2771 				md = m_getptr(m, npbytes -
2772 				    sizeof(struct mldv2_record), &off);
2773 				pmr = (struct mldv2_record *)(mtod(md,
2774 				    uint8_t *) + off);
2775 			} else {
2776 				/* current packet; offset from last append */
2777 				md = m_last(m);
2778 				pmr = (struct mldv2_record *)(mtod(md,
2779 				    uint8_t *) + md->m_len -
2780 				    sizeof(struct mldv2_record));
2781 			}
2782 			/*
2783 			 * Begin walking the tree for this record type
2784 			 * pass, or continue from where we left off
2785 			 * previously if we had to allocate a new packet.
2786 			 * Only report deltas in-mode at t1.
2787 			 * We need not report included sources as allowed
2788 			 * if we are in inclusive mode on the group,
2789 			 * however the converse is not true.
2790 			 */
2791 			rsrcs = 0;
2792 			if (nims == NULL) {
2793 				nims = RB_MIN(ip6_msource_tree,
2794 				    &inm->in6m_srcs);
2795 			}
2796 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2797 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2798 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2799 				now = im6s_get_mode(inm, ims, 1);
2800 				then = im6s_get_mode(inm, ims, 0);
2801 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2802 				    __func__, then, now);
2803 				if (now == then) {
2804 					CTR1(KTR_MLD,
2805 					    "%s: skip unchanged", __func__);
2806 					continue;
2807 				}
2808 				if (mode == MCAST_EXCLUDE &&
2809 				    now == MCAST_INCLUDE) {
2810 					CTR1(KTR_MLD,
2811 					    "%s: skip IN src on EX group",
2812 					    __func__);
2813 					continue;
2814 				}
2815 				nrt = (rectype_t)now;
2816 				if (nrt == REC_NONE)
2817 					nrt = (rectype_t)(~mode & REC_FULL);
2818 				if (schanged++ == 0) {
2819 					crt = nrt;
2820 				} else if (crt != nrt)
2821 					continue;
2822 				if (!m_append(m, sizeof(struct in6_addr),
2823 				    (void *)&ims->im6s_addr)) {
2824 					if (m != m0)
2825 						m_freem(m);
2826 					CTR1(KTR_MLD,
2827 					    "%s: m_append() failed", __func__);
2828 					return (-ENOMEM);
2829 				}
2830 				nallow += !!(crt == REC_ALLOW);
2831 				nblock += !!(crt == REC_BLOCK);
2832 				if (++rsrcs == m0srcs)
2833 					break;
2834 			}
2835 			/*
2836 			 * If we did not append any tree nodes on this
2837 			 * pass, back out of allocations.
2838 			 */
2839 			if (rsrcs == 0) {
2840 				npbytes -= sizeof(struct mldv2_record);
2841 				if (m != m0) {
2842 					CTR1(KTR_MLD,
2843 					    "%s: m_free(m)", __func__);
2844 					m_freem(m);
2845 				} else {
2846 					CTR1(KTR_MLD,
2847 					    "%s: m_adj(m, -mr)", __func__);
2848 					m_adj(m, -((int)sizeof(
2849 					    struct mldv2_record)));
2850 				}
2851 				continue;
2852 			}
2853 			npbytes += (rsrcs * sizeof(struct in6_addr));
2854 			if (crt == REC_ALLOW)
2855 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2856 			else if (crt == REC_BLOCK)
2857 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2858 			pmr->mr_numsrc = htons(rsrcs);
2859 			/*
2860 			 * Count the new group record, and enqueue this
2861 			 * packet if it wasn't already queued.
2862 			 */
2863 			m->m_pkthdr.PH_vt.vt_nrecs++;
2864 			if (m != m0)
2865 				mbufq_enqueue(mq, m);
2866 			nbytes += npbytes;
2867 		} while (nims != NULL);
2868 		drt |= crt;
2869 		crt = (~crt & REC_FULL);
2870 	}
2871 
2872 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2873 	    nallow, nblock);
2874 
2875 	return (nbytes);
2876 }
2877 
2878 static int
2879 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2880 {
2881 	struct mbufq	*gq;
2882 	struct mbuf	*m;		/* pending state-change */
2883 	struct mbuf	*m0;		/* copy of pending state-change */
2884 	struct mbuf	*mt;		/* last state-change in packet */
2885 	int		 docopy, domerge;
2886 	u_int		 recslen;
2887 
2888 	docopy = 0;
2889 	domerge = 0;
2890 	recslen = 0;
2891 
2892 	IN6_MULTI_LIST_LOCK_ASSERT();
2893 	MLD_LOCK_ASSERT();
2894 
2895 	/*
2896 	 * If there are further pending retransmissions, make a writable
2897 	 * copy of each queued state-change message before merging.
2898 	 */
2899 	if (inm->in6m_scrv > 0)
2900 		docopy = 1;
2901 
2902 	gq = &inm->in6m_scq;
2903 #ifdef KTR
2904 	if (mbufq_first(gq) == NULL) {
2905 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2906 		    __func__, inm);
2907 	}
2908 #endif
2909 
2910 	m = mbufq_first(gq);
2911 	while (m != NULL) {
2912 		/*
2913 		 * Only merge the report into the current packet if
2914 		 * there is sufficient space to do so; an MLDv2 report
2915 		 * packet may only contain 65,535 group records.
2916 		 * Always use a simple mbuf chain concatentation to do this,
2917 		 * as large state changes for single groups may have
2918 		 * allocated clusters.
2919 		 */
2920 		domerge = 0;
2921 		mt = mbufq_last(scq);
2922 		if (mt != NULL) {
2923 			recslen = m_length(m, NULL);
2924 
2925 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2926 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2927 			    MLD_V2_REPORT_MAXRECS) &&
2928 			    (mt->m_pkthdr.len + recslen <=
2929 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2930 				domerge = 1;
2931 		}
2932 
2933 		if (!domerge && mbufq_full(gq)) {
2934 			CTR2(KTR_MLD,
2935 			    "%s: outbound queue full, skipping whole packet %p",
2936 			    __func__, m);
2937 			mt = m->m_nextpkt;
2938 			if (!docopy)
2939 				m_freem(m);
2940 			m = mt;
2941 			continue;
2942 		}
2943 
2944 		if (!docopy) {
2945 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2946 			m0 = mbufq_dequeue(gq);
2947 			m = m0->m_nextpkt;
2948 		} else {
2949 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2950 			m0 = m_dup(m, M_NOWAIT);
2951 			if (m0 == NULL)
2952 				return (ENOMEM);
2953 			m0->m_nextpkt = NULL;
2954 			m = m->m_nextpkt;
2955 		}
2956 
2957 		if (!domerge) {
2958 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2959 			    __func__, m0, scq);
2960 			mbufq_enqueue(scq, m0);
2961 		} else {
2962 			struct mbuf *mtl;	/* last mbuf of packet mt */
2963 
2964 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2965 			    __func__, m0, mt);
2966 
2967 			mtl = m_last(mt);
2968 			m0->m_flags &= ~M_PKTHDR;
2969 			mt->m_pkthdr.len += recslen;
2970 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2971 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2972 
2973 			mtl->m_next = m0;
2974 		}
2975 	}
2976 
2977 	return (0);
2978 }
2979 
2980 /*
2981  * Respond to a pending MLDv2 General Query.
2982  */
2983 static void
2984 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2985 {
2986 	struct ifmultiaddr	*ifma;
2987 	struct ifnet		*ifp;
2988 	struct in6_multi	*inm;
2989 	int			 retval;
2990 
2991 	IN6_MULTI_LIST_LOCK_ASSERT();
2992 	MLD_LOCK_ASSERT();
2993 
2994 	KASSERT(mli->mli_version == MLD_VERSION_2,
2995 	    ("%s: called when version %d", __func__, mli->mli_version));
2996 
2997 	/*
2998 	 * Check that there are some packets queued. If so, send them first.
2999 	 * For large number of groups the reply to general query can take
3000 	 * many packets, we should finish sending them before starting of
3001 	 * queuing the new reply.
3002 	 */
3003 	if (mbufq_len(&mli->mli_gq) != 0)
3004 		goto send;
3005 
3006 	ifp = mli->mli_ifp;
3007 
3008 	IF_ADDR_RLOCK(ifp);
3009 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3010 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
3011 		    ifma->ifma_protospec == NULL)
3012 			continue;
3013 
3014 		inm = (struct in6_multi *)ifma->ifma_protospec;
3015 		KASSERT(ifp == inm->in6m_ifp,
3016 		    ("%s: inconsistent ifp", __func__));
3017 
3018 		switch (inm->in6m_state) {
3019 		case MLD_NOT_MEMBER:
3020 		case MLD_SILENT_MEMBER:
3021 			break;
3022 		case MLD_REPORTING_MEMBER:
3023 		case MLD_IDLE_MEMBER:
3024 		case MLD_LAZY_MEMBER:
3025 		case MLD_SLEEPING_MEMBER:
3026 		case MLD_AWAKENING_MEMBER:
3027 			inm->in6m_state = MLD_REPORTING_MEMBER;
3028 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3029 			    inm, 0, 0, 0, 0);
3030 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3031 			    __func__, retval);
3032 			break;
3033 		case MLD_G_QUERY_PENDING_MEMBER:
3034 		case MLD_SG_QUERY_PENDING_MEMBER:
3035 		case MLD_LEAVING_MEMBER:
3036 			break;
3037 		}
3038 	}
3039 	IF_ADDR_RUNLOCK(ifp);
3040 
3041 send:
3042 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3043 
3044 	/*
3045 	 * Slew transmission of bursts over 500ms intervals.
3046 	 */
3047 	if (mbufq_first(&mli->mli_gq) != NULL) {
3048 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3049 		    MLD_RESPONSE_BURST_INTERVAL);
3050 		V_interface_timers_running6 = 1;
3051 	}
3052 }
3053 
3054 /*
3055  * Transmit the next pending message in the output queue.
3056  *
3057  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3058  * MRT: Nothing needs to be done, as MLD traffic is always local to
3059  * a link and uses a link-scope multicast address.
3060  */
3061 static void
3062 mld_dispatch_packet(struct mbuf *m)
3063 {
3064 	struct ip6_moptions	 im6o;
3065 	struct ifnet		*ifp;
3066 	struct ifnet		*oifp;
3067 	struct mbuf		*m0;
3068 	struct mbuf		*md;
3069 	struct ip6_hdr		*ip6;
3070 	struct mld_hdr		*mld;
3071 	int			 error;
3072 	int			 off;
3073 	int			 type;
3074 	uint32_t		 ifindex;
3075 
3076 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3077 
3078 	/*
3079 	 * Set VNET image pointer from enqueued mbuf chain
3080 	 * before doing anything else. Whilst we use interface
3081 	 * indexes to guard against interface detach, they are
3082 	 * unique to each VIMAGE and must be retrieved.
3083 	 */
3084 	ifindex = mld_restore_context(m);
3085 
3086 	/*
3087 	 * Check if the ifnet still exists. This limits the scope of
3088 	 * any race in the absence of a global ifp lock for low cost
3089 	 * (an array lookup).
3090 	 */
3091 	ifp = ifnet_byindex(ifindex);
3092 	if (ifp == NULL) {
3093 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3094 		    __func__, m, ifindex);
3095 		m_freem(m);
3096 		IP6STAT_INC(ip6s_noroute);
3097 		goto out;
3098 	}
3099 
3100 	im6o.im6o_multicast_hlim  = 1;
3101 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3102 	im6o.im6o_multicast_ifp = ifp;
3103 
3104 	if (m->m_flags & M_MLDV1) {
3105 		m0 = m;
3106 	} else {
3107 		m0 = mld_v2_encap_report(ifp, m);
3108 		if (m0 == NULL) {
3109 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3110 			IP6STAT_INC(ip6s_odropped);
3111 			goto out;
3112 		}
3113 	}
3114 
3115 	mld_scrub_context(m0);
3116 	m_clrprotoflags(m);
3117 	m0->m_pkthdr.rcvif = V_loif;
3118 
3119 	ip6 = mtod(m0, struct ip6_hdr *);
3120 #if 0
3121 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3122 #else
3123 	/*
3124 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3125 	 * occur if we called in6_setscope() at transmission.
3126 	 * See comments at top of file.
3127 	 */
3128 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3129 #endif
3130 
3131 	/*
3132 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3133 	 * so we can bump the stats.
3134 	 */
3135 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3136 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3137 	type = mld->mld_type;
3138 
3139 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3140 	    &oifp, NULL);
3141 	if (error) {
3142 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3143 		goto out;
3144 	}
3145 	ICMP6STAT_INC(icp6s_outhist[type]);
3146 	if (oifp != NULL) {
3147 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3148 		switch (type) {
3149 		case MLD_LISTENER_REPORT:
3150 		case MLDV2_LISTENER_REPORT:
3151 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3152 			break;
3153 		case MLD_LISTENER_DONE:
3154 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3155 			break;
3156 		}
3157 	}
3158 out:
3159 	return;
3160 }
3161 
3162 /*
3163  * Encapsulate an MLDv2 report.
3164  *
3165  * KAME IPv6 requires that hop-by-hop options be passed separately,
3166  * and that the IPv6 header be prepended in a separate mbuf.
3167  *
3168  * Returns a pointer to the new mbuf chain head, or NULL if the
3169  * allocation failed.
3170  */
3171 static struct mbuf *
3172 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3173 {
3174 	struct mbuf		*mh;
3175 	struct mldv2_report	*mld;
3176 	struct ip6_hdr		*ip6;
3177 	struct in6_ifaddr	*ia;
3178 	int			 mldreclen;
3179 
3180 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3181 	KASSERT((m->m_flags & M_PKTHDR),
3182 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3183 
3184 	/*
3185 	 * RFC3590: OK to send as :: or tentative during DAD.
3186 	 */
3187 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3188 	if (ia == NULL)
3189 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3190 
3191 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3192 	if (mh == NULL) {
3193 		if (ia != NULL)
3194 			ifa_free(&ia->ia_ifa);
3195 		m_freem(m);
3196 		return (NULL);
3197 	}
3198 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3199 
3200 	mldreclen = m_length(m, NULL);
3201 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3202 
3203 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3204 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3205 	    sizeof(struct mldv2_report) + mldreclen;
3206 
3207 	ip6 = mtod(mh, struct ip6_hdr *);
3208 	ip6->ip6_flow = 0;
3209 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3210 	ip6->ip6_vfc |= IPV6_VERSION;
3211 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3212 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3213 	if (ia != NULL)
3214 		ifa_free(&ia->ia_ifa);
3215 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3216 	/* scope ID will be set in netisr */
3217 
3218 	mld = (struct mldv2_report *)(ip6 + 1);
3219 	mld->mld_type = MLDV2_LISTENER_REPORT;
3220 	mld->mld_code = 0;
3221 	mld->mld_cksum = 0;
3222 	mld->mld_v2_reserved = 0;
3223 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3224 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3225 
3226 	mh->m_next = m;
3227 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3228 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3229 	return (mh);
3230 }
3231 
3232 #ifdef KTR
3233 static char *
3234 mld_rec_type_to_str(const int type)
3235 {
3236 
3237 	switch (type) {
3238 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3239 			return "TO_EX";
3240 			break;
3241 		case MLD_CHANGE_TO_INCLUDE_MODE:
3242 			return "TO_IN";
3243 			break;
3244 		case MLD_MODE_IS_EXCLUDE:
3245 			return "MODE_EX";
3246 			break;
3247 		case MLD_MODE_IS_INCLUDE:
3248 			return "MODE_IN";
3249 			break;
3250 		case MLD_ALLOW_NEW_SOURCES:
3251 			return "ALLOW_NEW";
3252 			break;
3253 		case MLD_BLOCK_OLD_SOURCES:
3254 			return "BLOCK_OLD";
3255 			break;
3256 		default:
3257 			break;
3258 	}
3259 	return "unknown";
3260 }
3261 #endif
3262 
3263 static void
3264 mld_init(void *unused __unused)
3265 {
3266 
3267 	CTR1(KTR_MLD, "%s: initializing", __func__);
3268 	MLD_LOCK_INIT();
3269 
3270 	ip6_initpktopts(&mld_po);
3271 	mld_po.ip6po_hlim = 1;
3272 	mld_po.ip6po_hbh = &mld_ra.hbh;
3273 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3274 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3275 }
3276 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3277 
3278 static void
3279 mld_uninit(void *unused __unused)
3280 {
3281 
3282 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3283 	MLD_LOCK_DESTROY();
3284 }
3285 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3286 
3287 static void
3288 vnet_mld_init(const void *unused __unused)
3289 {
3290 
3291 	CTR1(KTR_MLD, "%s: initializing", __func__);
3292 
3293 	LIST_INIT(&V_mli_head);
3294 }
3295 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3296     NULL);
3297 
3298 static void
3299 vnet_mld_uninit(const void *unused __unused)
3300 {
3301 
3302 	/* This can happen if we shutdown the network stack. */
3303 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3304 }
3305 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3306     NULL);
3307 
3308 static int
3309 mld_modevent(module_t mod, int type, void *unused __unused)
3310 {
3311 
3312     switch (type) {
3313     case MOD_LOAD:
3314     case MOD_UNLOAD:
3315 	break;
3316     default:
3317 	return (EOPNOTSUPP);
3318     }
3319     return (0);
3320 }
3321 
3322 static moduledata_t mld_mod = {
3323     "mld",
3324     mld_modevent,
3325     0
3326 };
3327 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3328